Synesthesia: When Numbers Have Colors and Memory Gets an Extra Handle

Introduction

In the early 2000s a man described to researchers something he had lived with since childhood. Letters had colors. A was red. B was orange. C was yellow. He had never chosen the pairings and could not switch them off, and tested and retested months apart the colors came back the same every time. That consistency is the standard evidence that somebody is a synesthete rather than someone with a vivid imagination.

Then the researchers found where his alphabet came from.

His colors matched a specific set of Fisher-Price refrigerator magnets sold in the 1970s. Not approximately. Letter for letter [1]. And when he later learned Cyrillic, the colors followed, assigned by how much each new letter looked or sounded like an English one he already had. The Cyrillic characters most resembling the magnet-set capitals even came out more saturated.

That single case, one person, published in 2006, sits underneath almost every argument here. If a synesthete's colors trace to a toy, what is synesthesia? Wiring you are born with? A childhood memory practised until it became automatic? Something in between that nobody has cleanly described?

And there is a second question sitting next to it, the one most people actually search for. Synesthetes are supposed to have better memories. Almost every article says so. The primary research says something narrower, smaller and more useful. This piece is about both, and about what happens when you check the popular version of a science story against the studies underneath.

What Synesthesia Is, and What It Is Not

Synesthesia is an involuntary crossing of senses or concepts. One thing reliably triggers a second experience that nobody else gets from the same input. A letter brings a color. A sound brings a color. A number sits at a fixed spot in space in front of you. A word has a taste.

"Involuntary" is doing the heavy lifting. Synesthetes do not decide that A is red the way you might decide Mondays feel grey. The color arrives with the letter and stays attached to it for decades, though not perfectly.

It is not a disease. Nothing to treat, nothing to cure, nothing that gets worse. Much medical-style writing gets this wrong by borrowing the vocabulary of symptoms and diagnosis for something closer to left-handedness.

What separates synesthesia from ordinary association is consistency over time. Most people asked to pick a color for A will happily pick one, and a different one later, because they invent it each time. A synesthete picks the same shade, repeatedly, without remembering the last one. That is the whole basis of testing.

David Eagleman and colleagues turned that insight into a standardised instrument in 2007, the Synesthesia Battery, which shows each letter and digit three times in shuffled order and measures how tightly the color choices cluster [2]. The obvious worry is that one sitting cannot substitute for retesting months apart, so somebody checked. The battery returned a grapheme-color prevalence of 1.2 percent, in line with long-term retesting, which is about as good a validation as this kind of instrument gets [3].

The definition still has soft edges, and the people who built it say so. Answering two published commentaries, Julia Simner set out three problems with the existing criteria: they treat synesthesia as far more homogeneous than it is, it probably has several different neurological causes, and synesthetes and non-synesthetes may sit on a continuum rather than in two boxes [4]. A 2025 paper showed how the criteria interact in practice, with self-report and test-retest consistency pulling in different directions and classifying partly different people [5].

Two more findings complicate the picture in useful ways. Consistency and the strength of an association turn out to be separable things, so a person can have a rock-solid pairing that feels faint, or a vivid one that drifts [6]. And consistency depends on how familiar the grapheme is in the first place, which matters because familiarity is something experience builds [7]. And the pairings drift: their quality and consistency change with age rather than being frozen at some early point [8]. Stable enough to diagnose somebody decades apart is not the same as fixed.

None of this makes synesthesia less real. It makes the boundary fuzzier than the popular version admits, and the fuzziness is where the interesting research lives.

How You Prove Somebody Has It

Almost every article about synesthesia mentions that a test exists and then moves on. The design of that test is more interesting than the fact of it, and understanding it tells you what researchers actually think synesthesia is.

The problem is obvious once you state it. Synesthesia is private. Nobody can look inside your head and check whether A really came with red attached, or whether you just like red and letters. Self-report cannot settle it, because someone who wants the answer to be yes will find a color for every letter.

So you stop asking what color A is and start asking whether the answer holds still.

Here is the shape of it. You are shown a letter and given a full color picker, not a menu of eight options, and you choose the exact shade. Then another letter, then another, the whole alphabet and the digits in shuffled order, and then the whole set again, shuffled differently. And again. By the third pass you have no reliable memory of the shade you picked the first time, because nobody holds sixty exact colors in mind across a shuffled sequence.

Then the scoring compares your three answers for each character and measures how far apart they are in color space.

A synesthete's three reds for A land almost on top of each other. A non-synesthete asked to invent associations scatters, because the answers are generated fresh each time rather than read off something already there. The gap is wide enough to classify people, and validating the method against 2,847 randomly recruited participants returned a prevalence matching months-apart retesting [3]. None of it depends on believing anyone's account of their inner life.

That is a genuinely clever piece of design, and it is why the field could restart after a century of stalemate. It also explains why consistency, rather than vividness, became the operational definition. The test measures what it can measure and the definition follows the test, which means real synesthetic experiences that happen to be inconsistent will fail it. The field knows this.

The Field Took a Hundred Years to Get Serious

The history matters here, because the long gap in the middle explains why so much of what circulates about synesthesia is old folklore rather than measurement.

The century of near-silence after Galton is not an accident. Synesthesia depends on self-report, and for most of the twentieth century self-report was what experimental psychology was trying to escape. There was no way to check whether somebody was describing an experience or a metaphor. The consistency test solved that.

The Number Everybody Quotes and Nobody Explains

Open any consumer page about synesthesia and you will find a percentage. Usually four percent. Occasionally three. Almost never with a study attached, and never with the thing that matters, which is how the number was obtained.

Simner and colleagues state the problem in their own opening. Conventional wisdom held that synesthesia occurred extremely rarely, in about 0.05 percent of births, and that it affected women more than men. Theirs was the first prevalence test with sampling that did not rely on self-referral, using objective tests to establish that each case was genuine. The rate came back eighty-eight times higher [9].

Count only people who wrote in to researchers and you count the ones who noticed, had heard the word, and felt like writing a letter. That is where your familiar four percent was hiding, and it is why the method matters more than the number.

The same study delivered a second result that page-one search results still have not absorbed. The most common variant was not colored letters. It was colored days of the week. Grapheme-color synesthesia, the type nearly every article treats as the default, was less common than the type nobody writes about.

Then there is sequence-space synesthesia, where numbers, months or years occupy fixed positions in the space around you. Using a spatial consistency test alongside a new questionnaire on naive samples, one team estimated its prevalence at 8.1 percent in one study and 12.8 percent in another, and validated the classification against lab tasks [10]. Those people showed a spatial interference effect, better detection of faint visual stimuli, more reported use of visual imagery, and improved memory for certain public events.

Read that back. Depending on the variant, between one in twelve and one in eight people may qualify. Synesthesia is not rare. The rare part is noticing it.

Prevalence also moves with what you were asked to learn. The largest survey to date compared native speakers of transparent orthographies against speakers of opaque ones, expecting harder-to-learn languages to produce more synesthetes. The first pass went the other way, and exploratory analysis found the reason: learning a non-native second language was strongly associated with synesthesia [11]. Musicians show elevated rates too [12], and a 2026 survey of medical students found awareness of synesthesia was low, with its authors stating outright that self-report figures do not establish prevalence [13].

Notice how much of that table says "not established". That is deliberate. The numbers that exist come from different methods on different populations, and putting a confident percentage next to every variant would be inventing precision that the literature does not have.

The Memory Claim, Tested Properly

Here is the sentence that appears on almost every page about synesthesia: people with it tend to have better memories.

It is true. It is also so vague that it is nearly useless, and it hides a genuinely interesting set of results underneath. When researchers actually went looking for the memory advantage with proper group designs and standardised instruments, what came back was smaller and weirder than the folklore.

Start with the study that found nothing. Nicolas Rothen and Beat Meier tested thirteen grapheme-color synesthetes against a control group matched for age, handedness, education and gender, on a visual search task and a memory test. There were no significant group differences on either. Not close. The effect sizes for visual search sat around Cohen's d of .19 to .32, which is small, and for episodic memory there was nothing to speak of [14]. Their conclusion was that synesthesia by itself does not produce a strong performance advantage, and that the spectacular results in the literature came mostly from case studies of unusual individuals. A year later the same lab came back with a different instrument. This time they used the Wechsler Memory Scale, a standardised clinical battery, on a group of grapheme-color synesthetes. Now there was an advantage, and it had a shape. The synesthetes did better on episodic memory. They did not do better on short-term memory. And the authors wrote the sentence that no consumer article will print: performance was still within the ordinary range [15].

Sit with that. The study built to measure the effect found the synesthetes inside the normal distribution. Better than average, yes. Extraordinary, no.

Other groups found the advantage too, and each one added a constraint. Nine color-graphemic synesthetes tested on both synesthesia-triggering verbal tasks and non-triggering visuospatial tasks showed an advantage on both, and the advantage was located primarily in the initial encoding of the material rather than in how well it survived afterwards [16].

Two things there pull in opposite directions. Encoding rather than storage is a clue about mechanism, and this section builds on it. But an advantage on visuospatial tasks that trigger no color at all is hard to fit into a story about colors helping. Hold onto that one. It comes back.

Two things jump out of that table. The first is that no single study shows a large general memory advantage. The second is that the studies do not contradict each other so much as narrow each other, and the narrowing points in one direction.

The same lab publishing a null and then a positive result looks like a contradiction and is not. The 2009 study used a visual search task and a memory test with thirteen people. The 2010 study used a clinical battery that separates memory into subscales, and a test that lumps episodic and short-term memory together will wash out an effect living in only one of them. The second study resolved the first by looking where the effect could show up.

It is also a warning about summaries. Pick the 2009 abstract and you can write "synesthetes have no memory advantage". Pick the 2010 one and you can write "synesthetes have superior episodic memory". Both are defensible from a single paper. Neither is what the pair says.

What the Advantage Is Actually Made Of

The clearest result on this question comes from a recognition memory experiment that was designed to break the effect apart.

Jamie Pritchard and colleagues built stimuli out of three features at once: a shape, a color and a location. Then they tested recognition using distractor items in which exactly one of those features had been unbound and swapped. Shape A with color B in location A. Shape A with color A in location C. If synesthetes simply have better memories, they should catch all three kinds of swap more often.

They did not. Synesthetes had higher recognition rates overall, and their false alarm rates were lower only when color was the unbound feature. Not shape. Not location [17].

That is a very specific result and it changes the whole story. The advantage is not a better memory. It is a better color channel, feeding into memory. Terhune and colleagues found the same shape of thing in visual working memory, where the enhancement was dimension-specific rather than general [18]. Ward and colleagues found enhanced recognition memory across several categories of material in grapheme-color synesthesia, which fits, because most material a literate person encounters passes through letters and digits on its way in [19].

The field's own review of all this refuses to let the tidy version stand. Richer encoding of the stimuli that trigger a color is the obvious explanation, and the reviewers say plainly that it is unlikely to be the whole one. Not everything that triggers synesthesia is remembered better. Digit span is not. And some things that trigger no color at all are remembered better anyway. Their summary is blunter still: synesthetes tend to have better visual memory than verbal memory, which is the wrong way round if colored letters were doing all the work [20].

Hold that against the color-binding result and you get the honest position. The advantage is sharpest where a color is involved and it is not sealed inside that boundary. Something broader is happening at the join between perception and memory, and the color is its most visible part rather than the whole.

A systematic comparison then asked the blunt version of the question: are synesthetes exceptional beyond their associations at all, across creativity, personality and cognition? The differences it found were smaller than those reported in the literature, and the authors suggested earlier studies may have overestimated them through biased recruitment. Most of the effects did replicate. They were just modest [21]. There is one more result that is easy to miss and hard to unsee. Single-case studies of extraordinary synesthetic memory do exist, including one showing that a synesthete's photisms affected name recall [22]. One person. That is not a criticism of the study, which was careful. It is a caution about what happens when a case study gets summarised into a general claim by the time it reaches a magazine.

One Hour, Then One Year

The single best piece of evidence for the extra handle came from a study that simply waited.

Katrin Lunke and Beat Meier recruited people with four different types of synesthesia, each with different inducer and concurrent pairings, and had them learn three kinds of material: things related to their trigger, things related to their synesthetic experience, and things unrelated to their synesthesia at all. Music, words and colors. Then they tested recognition after one hour, and again after one year [23].

After an hour, two groups showed synesthesia-specific advantages: the grapheme-color synesthetes, and those with both grapheme-color and sound-color.

After a year, only the grapheme-color group still did.

A year is a long time to hold a cohort together, and the result is worth more than a dozen short experiments. The benefit is not a general encoding boost fading evenly. It is tied to one kind of pairing, and that pairing keeps paying out after the others stop.

Think about what the diagram is describing. An ordinary reader stores a word as a word. A grapheme-color synesthete stores the word and, alongside it, a color pattern generated automatically by the letters. Two routes in. Two routes back out. When one fails, the other can still deliver.

That is the extra handle, and it is not a metaphor invented for a headline. It is what the pattern of results looks like from the outside.

There is a reason the advantage concentrates at encoding, and the recognition result explains it. If the color arrives automatically the moment you first see the word, it is part of what gets stored, at no attentional cost. A free extra feature bound into the trace. At retrieval the color can only help if the thing you are reaching for had one. Shapes did not benefit. Locations did not benefit. Colors did.

That is where the effect is sharpest. It is not where the effect ends, and the review above is the reason to keep that distinction in view.

You can see why a year later only one group was still ahead. Letters and digits are involved in almost everything a literate adult learns. A sound-color pairing fires far less often, gets less practice, and has fewer occasions to be reinforced across twelve months.

You Can Build One Deliberately

Here is the part that transfers to you, whether or not you see colors.

The mechanism proposed for the synesthete's advantage is not exotic. Extra perceptual experience gives richer encoding and more ways back in, which is the explanation the field reaches for first and the one its own review calls incomplete rather than wrong [20]. A second code, attached to the same material, available when the first one stalls. The entire craft of mnemonics is built on manufacturing exactly that, deliberately.

If you have ever used a memory palace, you have done this. You take material with no spatial structure and force a spatial structure onto it, so that a walk through a familiar building becomes a second retrieval route into the same information. Our piece on the method of loci and why placing beats repeating covers how that works and why the placing step matters more than the repetition.

Chunking does something related. It does not add a sensory channel, it changes the unit, so that seven items become three. If you want the mechanics of that, our article on chunking takes it apart.

The difference between the synesthete and you is not the mechanism. It is the cost. Their second code arrives free and automatic. Yours has to be built, and rebuilt, and it decays if you stop. What that cost buys, and whether a deliberately built code can match an involuntary one, is not something the studies in this article measured, so I am not going to tell you it can. What they do show is that the involuntary version produces an advantage that stays modest.

None of which means you should try to become a synesthete. We will get to the training studies later, and what they show turns out to depend entirely on what you think the word covers.

Inside the Brain, and the Argument About What Is There

The mechanistic story most people have heard goes like this. In grapheme-color synesthesia the brain region that recognises letters sits right next to a color region called V4, and in synesthetes the two are connected more than they should be, so activating one leaks into the other. That account was laid out in 2005 and it is still the standard reference point [24]. The first strong physical evidence came from Romke Rouw and Steven Scholte, who put eighteen grapheme-color synesthetes through diffusion tensor imaging and compared them against matched controls. The synesthetes showed greater anisotropic diffusion, which indicates more coherent white matter [25].

The interesting part was what the connectivity predicted. Synesthetes divide roughly into two groups by their own description. Projectors see the color out in the world, on the page, where the letter is. Associators see it in the mind's eye.

That distinction is worth pausing on, because it is the thing most non-synesthetes get wrong when they try to imagine the experience. The common assumption is that everybody with grapheme-color synesthesia sees a printed black 7 turn visibly green on the page. Many do not. For an associator the color is closer to known than to seen, in the way you know a friend's voice without hearing it, present and specific and located somewhere that is not quite the visual field. Projector accounts are the ones that make it into magazines, because seeing a color out in the world is the more dramatic story to tell.

Greater connectivity in the inferior temporal cortex was particularly strong in projectors, and did not separate the subtypes elsewhere [25]. A subjective difference in how the experience feels lined up with a structural difference in where the wiring was heaviest.

If you want one detail from this literature to keep, make it this one.

The follow-up scanned forty-two synesthetes, sixteen projectors and twenty-six associators. All shared machinery in posterior superior parietal lobe, which integrates sensory information. Beyond that they diverged: the out-in-the-world experience tracked areas for perceiving and acting, the in-the-mind experience tracked hippocampus and parahippocampal gyrus [26]. Memory structures. Synesthetes who describe the color as known rather than seen are running it partly on the hardware that stores things. Another team showed effective connectivity rather than structure determines which of the two you get [27].

Timing evidence came next, and it did not settle the way the cross-activation story needed. Magnetoencephalography, which resolves activity in milliseconds, showed early activation of V4 in grapheme-color synesthetes [28]. Early matters: if the color region fires late, the color could be a thought about the letter, and if it fires early it is closer to part of seeing it. Then MEG was run on six associator synesthetes. Activity distinguishing color-inducing from non-inducing letters showed up at about 190 milliseconds, in extrastriate visual cortex and superior parietal lobe, and not in controls. The authors read that as too late for straight cross-activation and a better fit for a rival account where the color arrives through feedback released from inhibition [29].

So the timing evidence does not point one way. It points at two accounts that have been arguing for twenty years, and if you were told the cross-activation picture was settled, you were told a simplification.

A pre-registered study then found projectors detect their own photisms at lower thresholds than associators, while the two groups are identical on real colors [30].

Timing is the pressure point for the whole cross-activation account, and it is why so much effort has gone into measuring it. If the color region activates late enough, the color could be something the brain adds after it has already identified the letter, which would make synesthesia a conceptual event dressed up as a perceptual one. If it activates early, that story is much harder to tell. Millisecond resolution is not a technical detail here. It is the argument.

Then somebody broke it. Using transcranial magnetic stimulation, researchers disrupted the automatic integration of synesthetic color and grapheme from outside the skull [31]. The binding is not a fixed property of the person. It is an ongoing process that can be interfered with, which is about as direct a piece of causal evidence as this field gets.

Resting-state work added another layer: intrinsic network connectivity, measured while participants were doing nothing at all, tracked how consistent their synesthetic experiences were [32]. And a case that constrains any purely sensory account: a congenitally blind synesthete showed activation of color-selective areas of visual cortex [33]. Colors in a brain that has never seen one.

The Review That Said Most of It Was Unsupported

Now the part that consumer articles leave out entirely.

In 2015 Jean-Michel Hupé and Michel Dojat reviewed every study that had searched for the neural correlates of synesthesia, along with the structural differences claimed for synesthetes. Their conclusion was blunt. Most of the claimed differences are unsupported, and the reasons are low statistical power, statistical errors and methodological limitations [34].

They went further. Rather than a brain anomaly, they suggested synesthesia could be reconsidered as a special kind of childhood memory.

Hold that thought against the magnets. It fits uncomfortably well.

The same group had already published a study finding that the neural bases of grapheme-color synesthesia are not localised in real color areas [35], which cuts directly against the neat cross-activation diagram everybody draws.

That is what a live disagreement looks like, and you should see it rather than a smoothed version. One side has striking imaging results. The other says many were underpowered. Neither is being unreasonable.

Then a Pre-Registered Study Pushed Back

The reply came in 2024, and it was built to answer exactly that criticism.

Jamie Ward and colleagues ran a pre-registered study, which means the analysis plan was locked before the data were seen, closing the door on the flexible-analysis problem that underpowered imaging studies are accused of. They assessed thirteen different brain-based biomarkers, structural and functional, for classifying synesthetes against general population samples, using machine learning models built on subject-specific cortical parcellations [36].

All thirteen performed above chance. The strongest single predictor was intracortical myelin, which had never been implicated in synesthesia before.

A companion paper answers the critique a different way, and its title says the quiet part out loud: when small effect sizes become huge. The argument is statistical. Psychology learned to expect small effects because it measures one variable at a time, and a multivariate effect size can turn a collection of small differences into a very large one [37]. On that reading Hupe and Dojat were right that no single measure separates synesthetes cleanly, and wrong to conclude they are barely different. The electrophysiology tested the other obvious objection. Perhaps the differences belong not to synesthesia but to atypical sensory sensitivity, which synesthetes and autistic people share and which varies widely in everyone else. Synesthetes showed a modulated visual evoked potential from fusiform cortex, a larger auditory response, and altered induced power, and none of it was predicted by sensory sensitivity [38].

Notice what you are watching. Each objection to the imaging evidence gets a study built to answer it. Slower than a headline, and the only thing that settles anything.

Two newer studies keep the pattern going. Blood levels of a protein involved in growing and protecting neurons came out higher in grapheme-color synesthetes than in matched controls [39]. And resting-state work found projectors and associators showing distinct but overlapping patterns, in regions tied to memory and executive function rather than to color [40]. Projectors and associators keep coming apart wherever anyone looks, which is the strongest hint here that the field has been averaging over two different things.

So where does that leave the question? Honestly, unresolved, but moving. The pre-registered evidence is stronger than what Hupé and Dojat were reviewing. Their criticism was about the state of the literature in 2015, and the literature changed partly because of it. That is the system working, slowly and in public.

Where the Colors Come From, Part One: The Genes

Synesthesia runs in families. Familiality studies have found evidence of a strong genetic predisposition, and it is one of the few things in this field that is not really argued about [41]. Turning that into genetics has been much harder than expected. Julian Asher and colleagues ran a whole-genome linkage scan for auditory-visual synesthesia across forty-three families, 196 people, using 410 microsatellite markers. They found genome-wide significant linkage to one region of chromosome 2 and suggestive linkage to three others [41].

Complex means no single gene. It means many contributions, probably interacting, probably different across families.

Four candidate regions, not one gene you could point at and name. That is the honest version, and a long way from what you will read elsewhere.

The same paper killed the tidiest story in the field. Early pedigree work supported single-gene X-linked dominant inheritance, resting on a skewed sex ratio and an apparent absence of father-to-son transmission. Asher's scan found no support for X linkage at all, and identified two confirmed cases of a father passing synesthesia to a son [41].

One counterexample ends an X-linked dominant hypothesis. They had two. What replaced it is messier: an oligogenic trait, several genes contributing, more than one mode of inheritance, different loci in different families.

Amanda Tilot and colleagues came at it from the other end. Instead of scanning many families for a shared signal, they took three families with sound-color synesthesia running through three or more generations and sequenced their exomes. In each family they found rare genetic variants that fully cosegregated with the trait, and across the three families the candidates converged on thirty-seven genes enriched for axonogenesis, the process by which growing neurons send out their connections [42].

Different genes in different families, pointing at the same developmental process. That is a much more plausible shape for this trait than one gene.

What gets inherited may not be the colors at all. Synesthesia is linked to a distinctive and heritable cognitive profile, a cluster of tendencies rather than a single trait [43]. The twin work sharpens that. A co-twin control study of sixty-five twins, including twenty-two complete pairs, looked at how the degree of grapheme-color association relates to autistic traits and perceptual style within each pair [44]. Comparing twins to each other rather than to strangers strips out everything the two of them shared growing up, which is the sort of design this question badly needed.

The associations were stronger within pairs than across the sample, which is the design working. The unexpected part: the twin with more grapheme-color association was better at recognising fragmented images, where a whole must be assembled from broken pieces. That cuts against the picture of synesthetes as unusually detail-focused, and the authors suspect memory or imagery rather than attention.

Families carrying synesthesia also carry more autism than expected, and no more schizophrenia than expected [45]. Twin modelling then moved the centre. Self-reported synesthesia correlated with features of autism, ADHD, obsessive-compulsive disorder, anxiety, depression, psychotic-like experiences, eating disorders and hypomania. The strongest link was not autism. It was obsessive-compulsive features, at 0.28, with genetic factors accounting for more than half of most associations and shared family environment for essentially none [46].

And there is a standing question underneath all of it. If synesthesia is heterogeneous and polygenic, as the genetics suggests, then either it once conferred some advantage or it is a side effect of something else that did. Nobody has been able to tell which, and the honest position is that working out the genetics is the route to answering it rather than the answer [47].

So the genetic case is real and it is not simple. Nowhere in any of it is there a number like "forty percent of cases are genetic", which several consumer health pages print without a source. There is no study behind that figure that I could find, and it does not appear anywhere in the source pool for this article. It should not be repeated, and it is not repeated here.

Where the Colors Come From, Part Two: The Magnets Get Bigger

Now back to the man with the Fisher-Price alphabet.

A single case is a single case. What made it an argument was scale. The same researchers found eleven synesthetes whose pairings were startlingly similar and traceable to childhood toys containing colored letters, the first data showing learned synesthesia in more than one person [48].

Their framing is the interesting part. They argued synesthesia can be thought of as the automatic retrieval of highly specific mnemonic associations, with perceptual content brought to mind in a way similar to mental imagery. Not a wire crossing. A memory practised until it stopped feeling like retrieval.

The obvious objection is that a synesthete might see the toy, notice the match, and reconstruct the memory backwards. The birth-cohort data is what answers it, because a person cannot retrofit a match to a product that did not exist when they were four.

Then they went bigger. Across a large sample of English-speaking grapheme-color synesthetes they estimated the minimum proportion whose matches came from an external source. At least six percent, four hundred of 6,588, had learned many of theirs from one widely available colored letter toy [49].

The birth-cohort pattern is the part that makes it hard to argue with.

Among synesthetes born five or more years before the toy was manufactured, none had colors matching it, while among those born in the decade after it launched the proportion approaches fifteen percent for some five-year windows [49]. And when the researchers compared the two groups on everything else, the only difference between synesthetes with toy-matched colors and synesthetes without was exposure to the toy. There is a whole theoretical position built on results like these, which treats learning as the organising principle of the entire literature rather than a footnote to it [50]. On that account the reason synesthetes remember well is not that synesthesia hands them a bonus. It is that synesthesia is itself a product of unusually strong learning, and the memory advantage is a symptom of the same underlying tendency rather than a consequence of the colors.

That is the callback. The awkward result was the nine synesthetes who did better on visuospatial tasks triggering no color at all. Under the colors-help-memory story it is a loose end. Under this one it is the point: if what these people have is an unusual capacity to build and hold associations, the colors and the memory scores are two outputs of one thing. Neither account explains everything. This one explains that.

Six percent is a minority, and nobody is claiming all synesthesia comes from refrigerator magnets. But it is not noise either. It is measured, cohort-dated and repeatable. Whatever synesthesia is, learning can supply its content in a substantial number of people.

The Result That Makes Both Stories Uncomfortable

If synesthesia were simply well-learned associations, you would predict that synesthetes are good at learning new grapheme-color pairings. They have the machinery. They have practice.

They are worse.

That single fact has done more to slow this argument down than anything else in the literature, because it is the one result neither camp can absorb without amendment.

David Brang and colleagues tested exactly this and found impaired acquisition of novel grapheme-color correspondences in synesthetes [51]. Their existing associations appear to interfere with new ones, which is not what a pure learning account predicts. It is worth separating two things that sound the same and are not. Learning an arbitrary new pairing, red for a letter that is already green, is the operation Brang found synesthetes are bad at. Extending an existing color to a new trigger is the operation the ideasthesia researchers describe continuing across a synesthete's life [52]. The first is overwriting. The second is spreading. A system can be resistant to one and open to the other.

The developmental data sit in between. Julia Simner and colleagues did the study this field needed: they followed real children over years rather than asking adults to remember their childhood. Over six hundred children were randomly sampled and tested at six or seven, then at seven or eight, then revisited three years later. What they found was a developmental progression. Synesthetic associations start as relatively chaotic pairings and mature over time into a fixed, consistent system [53].

Neither camp gets that cleanly. The associations are there early, which the learning-only account struggles with. They are not fixed early, which the born-fully-formed account struggles with. Something is being refined, over years, by experience.

There is a whole line of research that takes the middle position seriously and calls it ideasthesia. The claim is that what triggers the second experience is the meaning of the inducer, not its raw sensory form. On that account synesthesia keeps developing through childhood and keeps being enriched afterwards: existing colors get adopted by new triggers, and new colors get formed, throughout a synesthete's life [52]. The objection this raises to the standard model is pointed. Neurophysiological theories treat synesthesia as perceptual and therefore look at the brain regions that do perception, while a large body of behavioural work keeps finding that conceptual information is doing the work [54]. Those two literatures have been talking past each other for years.

The two camps rarely converge because one measures brains and the other measures behaviour, and the disagreement is not about what either measurement showed. It is about which one gets to define the phenomenon.

A 2022 experiment found a way to test it. If the color attaches to the written digit, it should arrive fast. If it attaches to the concept of the quantity, it should arrive later, because the quantity has to be worked out first. So the researchers showed the same number as a digit, as a dice face, and as a scattered dot pattern, then measured how much the synesthetic color sped up naming a matching color patch at delays from zero to eight hundred milliseconds [55].

Both pathways were there and behaved differently. The effect from the written digit arrived quickly and persisted. The effect from the quantity unfolded later and faded faster.

So neither camp gets the whole territory. The color attaches to the symbol and to the meaning, two mechanisms on different clocks inside what everybody treated as one thing.

Why Is A Red?

Ask a room of grapheme-color synesthetes what color A is and you will get more agreement than chance. Not unanimity. But a tilt, and studies using large populations of synesthetes have found that grapheme-color pairings are influenced by numerous properties of the graphemes themselves [56].

That tilt is a data source. If the pairings were purely idiosyncratic wiring, there should be no population-level pattern at all. If they were purely learned from one toy, everybody exposed to that toy would match and nobody else would. The truth is somewhere in the middle, and untangling it requires separating several things that normally travel together: the shape of the letter, the sound it makes, what it means, and where it sits in the alphabet.

Nicholas Root and colleagues did that with a multilingual dataset, which was the trick. In an English-only sample those explanations are hopelessly tangled, because the shape of a letter, the sound of its name and the words it starts are all fixed together. Their contribution was the method: the first demonstration that comparing across languages can pull previously confounded hypotheses apart and expose which influences are potentially universal [56]. The follow-up ran the comparison across Dutch, English, Greek, Japanese, Korean, Russian and Spanish, and the answer split cleanly in two [57]. Linguistic influences, meaning color words and semantic associations, varied significantly in size between languages. Y is yellow because of the English word, and that reason simply does not exist for a Korean synesthete. Non-linguistic influences, meaning shape-to-color associations, did not differ between languages at all.

So the answer to why A is red comes in two parts. Part of it is your language, which means part of it is learned. Part of it is not your language, which means part of it is something the human visual system does with shapes regardless of what you grew up speaking. Neither camp gets the whole answer, which by this point in the article should be starting to look like the pattern rather than the exception.

There is a related result from outside synesthesia that is worth knowing. In non-synesthetes, categorical consistency of parity and magnitude, meaning whether a number is odd or even and how big it is, makes implicit learning of color-number associations easier [58]. The ordinary brain already has a weak version of the structure that synesthesia expresses strongly. And the classic sound-symbolism finding, that people match the nonsense word maluma to a rounded shape and takete to a spiky one, turns out to develop later than the standard story claims. A longitudinal study found no evidence for the shape-sound effect emerging the way it was assumed to [59]. Cross-modal correspondences are not simply built in and waiting.

Can You Learn It?

You can learn associations. That much is not in doubt, and the training literature has been reviewed in full: non-synesthetes can be trained to the point where they show synesthesia-like behaviour on lab tasks [60].

Whether they have synesthesia is another question entirely.

Rothen and Meier came down further on the yes side than you might expect. Their starting point is that synesthetic associations are demonstrably not random and that specific cases trace to prior experience, so a learning component is built into how they form. Their conclusion, under an explicit working definition, is that synesthesia can potentially be learned given the right training [60]. A related paper works through what would even count as trained grapheme-color synesthesia, which is harder than it sounds once you write the criteria down [61].

The sharpest counterweight came from Ophelia Deroy and Charles Spence, whose paper is titled, with visible impatience, "Training, hypnosis, and drugs: artificial synaesthesia, or artificial paradises?" [62]. Their argument is that the various methods claimed to induce synesthesia produce something that resembles it on the outside and lacks the properties that define it: the automaticity, the involuntary quality, the decades-long stability.

The disagreement here is not really about the data. Both sides agree that trained participants start responding to letters in ways that resemble synesthetic behaviour. The argument is about what the word should cover. If synesthesia means a set of measurable effects on tasks, training produces it. If it means an involuntary experience that shows up unbidden in daily life and holds for thirty years, training has not been shown to produce that, and the studies are far too short to find out.

Psychedelics can produce transient cross-modal experiences and that literature has been reviewed [63]. It is a documented phenomenon and it is not the same thing as developmental synesthesia, which is present from childhood and consistent across a lifetime.

So the honest answer to "can I train myself into synesthesia" is that you can build associations behaving a bit like it under test conditions, and that it is not the same experience. Whether a trained version carries the memory benefit is a separate question these studies were not built to answer. The useful part was never the color anyway. It was the second code, and you can have that.

Maybe Not. Yes, But.

Sometimes an argument is legible from outside, and this is my favourite example in the field.

Sequence-space synesthetes see numbers, months and years laid out in space, so it seems obvious they would have better spatial imagery. In 2012 a paper tested it and answered in its own title: "Do sequence-space synaesthetes have better spatial imagery skills? Maybe not" [64]. Three years later another team ran it again and replied, also in the title: "Do sequence-space synaesthetes have better spatial imagery skills? Yes but there are individual differences" [65].

That is the shape of science in two titles. Somebody asks. Somebody answers no. Somebody looks harder, finds yes for some people and no for others, and the question gets sharper instead of closed.

The layouts have been imaged, and the scan said something the behavioural argument had not. Ten people whose number forms run left to right were compared against controls on two tasks. Judging which number was bigger produced no difference. Judging which came first produced extra activation in the intraparietal sulcus on both sides [66]. Number forms are about order, not quantity, which is why a test of spatial imagery was always going to answer messily.

The Neighbours

Synesthesia does not sit alone, and the reason to walk its neighbourhood is not to collect co-occurrences. It is that every neighbour turns out to be a different setting of a dial rather than a missing part, and that is the finding this whole section is for.

Start with the one that should not be possible. Aphantasia is the absence of voluntary visual imagery, and synesthesia is supposed to be imagery in surplus, so the two ought to be incompatible. They are not. Grapheme-color synesthesia can be objectively diagnosed in people with aphantasia, which means visual imagery is not required for synesthesia to happen at all. What aphantasia changes is the flavour: those synesthetes come out as associators rather than projectors, feeling the color rather than seeing it out in the world [67].

Read that twice. It removes a whole class of explanation: whatever binds a color to a letter is not the machinery that paints pictures in your head. For the other end of that spectrum, we have a piece on how memory works without pictures.

Autism is the other well-documented neighbour, and the number is larger than most people expect. In the first prevalence study to test the two conditions against each other, 164 autistic adults and 97 controls were screened, and synesthesia turned up in 18.9 percent of the autistic group against 7.22 percent of controls [68]. Almost three times the rate.

Three times is not a subtle overlap. What makes it hard to act on is that both are screened by questionnaire in large samples, and questionnaires reward people who notice things about themselves.

The comparison with schizophrenia is more interesting than a simple clustering question, because it was designed around a mechanism. Perception is treated as inference: what you see is sensory evidence weighted against what you already expect. If people differ in how much weight they give each side, that one dial could explain very different perceptual lives. Twenty synesthetes, twenty medicated schizophrenia patients and twenty-six controls rated how visible stimuli buried in noise appeared to them [69].

The two groups moved in opposite directions. The schizophrenia patients needed more sensory evidence before calling something visible, which fits an overreliance on the incoming signal. The synesthetes needed less, but only for stimuli that trigger their own synesthesia, which fits unusually precise long-term expectations about exactly that material.

One dial, turned two ways, and the synesthetes' setting is specific to the material they have spent a lifetime forming expectations about.

You can see why that framing is useful to you even if you have neither condition. It says perception is something your brain negotiates rather than receives, and the terms of that negotiation differ between people.

Faces are the odd one out. Synesthetes have been reporting trouble recognising faces for years, and the brain regions involved sit close to the ones implicated in grapheme-color synesthesia, so somebody tested it. Sixteen grapheme-color synesthetes and sixteen matched controls took a standard face memory test and a holistic face processing task [70].

The synesthetes were significantly worse at recognising faces, and the deficit came from a reduced upright advantage. Their accuracy on the holistic task was lower too, though the holistic advantage itself was intact. They also rated themselves as worse at faces, so the complaint that started the study was accurate.

Face-recognition difficulty is not unique to synesthetes, for what it is worth. In a survey of 115 people with self-reported face blindness, 57 percent reported at least one other developmental difficulty [71]. Whatever this is, it is a common neighbourhood rather than a private one. There is even a difference in stopping. Six grapheme-color synesthetes and twenty-four non-synesthetes ran visual, auditory and somatosensory Go/No-go tasks with event-related potentials recorded, and the synesthetes made more omission errors while some of their component latencies were shorter [72]. Six people is a very small sample and the result should be treated as a lead rather than a finding.

None of these is dramatic on its own. What they add up to is the same thing the memory review said from the other direction. The recognition experiments show where the memory advantage is sharpest, and it is sharpest around color. Everything in this section shows that whatever produces the color is not confined to color, because it also turns up in faces, in stopping, and in how sensory evidence gets weighted against expectation. A benefit with a sharp edge does not require a cause with one.

Perception is not uniformly enhanced either, which matters for the dial framing. Synesthesia for color is linked to better color perception and worse motion perception, a trade rather than a gift [73], and imagery advantages extend past the visual [74].

So the promise at the top of this section pays off. Imagery vividness runs from none to overwhelming, and synesthesia survives at the bottom of that range. Sensory expectation runs from loose to unusually precise, with synesthesia at one end and schizophrenia at the other. Color perception runs up while motion perception runs down. Nobody has found a switch that is simply off in most people and on in synesthetes, and the longer researchers look, the less likely it seems that a switch exists.

Ward has argued that synesthesia is best understood as an emergent feature of adaptive neurocognitive processes rather than a discrete condition sitting on top of an otherwise typical brain [75]. That framing does a lot of work, because it explains why the boundaries keep refusing to be crisp.

The Variants Nobody Writes About

Grapheme-color gets nearly all the coverage because it is easy to test. The other types are stranger, and each runs on machinery you already have. That is the thread to hold. None is an extra sense. Each is an ordinary system pushed somewhere ordinary systems do not go.

Chromesthesia is sound to color, the variant most people have heard of even without the word. Colored-music synesthetes show enhanced structural connectivity along the auditory-to-visual pathways you would expect [76], and a review found chromesthesia and absolute pitch have barely been studied together despite obvious overlap in how musicians describe them [77].

The ordinary-machinery point lands hardest here. Ask non-synesthetes what they see while listening to music and they report story-like visual imagery, some personal and some shared across listeners [78]. The line between a synesthetic and a non-synesthetic response to music is blurrier than the labels suggest. If you want more on that side of it, our article on music and memory goes into how sound anchors recall. Mirror-touch makes the same point with a different system. Seeing somebody else touched produces a felt sensation on your own body, which is your ordinary capacity to flinch at another person's injury turned up until it stops being figurative. It has been linked to empathy, and a screening tool exists for it [79].

Tickertape may be the strangest of the well-documented types. Spoken words appear as visible written text, as though the sentence you just heard were printing itself out. One man with it was scanned across a series of experiments, and the result is the cleanest illustration of this section's point: the areas more active in him than in controls during speech were the reading areas, including the visual word form area [80].

Reading turns letters into speech. This is the same network backwards, turning speech into letters. Not a new faculty. An existing one, pointed the other way. Our piece on the visual word form area covers the machinery being reversed.

Underneath all of them is a timing question. Is the second experience part of perception, or does it arrive just after? The whole cross-activation model depends on the answer. Early V4 activation says perception. The semantic results say something later. Our article on the quarter second of sensory memory covers the window this argument happens in.

What This Actually Means for Memory

Strip everything back and the useful finding is not about synesthesia at all.

The advantage is modest, sharpest around color, strongest at encoding, and durable over a year only for one variant. Each constraint says something about memory in general. An extra code helps. It helps most where the code applies. It helps at learning more than at recall. And it keeps helping if the code is stable.

That is a description of dual coding, and it applies to you exactly as much as it applies to somebody who sees a green seven.

It also puts the popular claim in its place. Extraordinary synesthetic memory shows up mostly in single-case reports, and against matched controls the effect shrinks to something ordinary. That pattern repeats across memory research. Our piece on whether photographic memory exists walks the same arc: a striking case, a general claim built on it, a much smaller real effect underneath.

None of this diminishes synesthesia. It is real, involuntary, consistent enough over decades to be diagnostic, and it leaves traces in white matter and resting-state networks. It just is not a superpower, and the people who have it have generally never said it was.

What Is Still Open

Three questions remain genuinely unsettled. Whether there is a reliable brain signature, with an underpowered-literature critique on one side and a pre-registered study on the other. Where the associations come from, with forty-three families and three sequenced pedigrees against six percent of a 6,588-person sample matching a toy. And how big the memory advantage is, which is the closest to settled: real, modest, sharpest around color, and never the thing a case study made it sound like.

If you take one thing from this, take the habit rather than the fact. When a claim about the mind reaches you already simplified, the interesting material is almost always in what got removed on the way. Take the percentage everybody quotes. It came from the first study that stopped waiting for volunteers to write in, it applies to the varieties that study could test, and the same paper found that the most-written-about variant is not the most common one. Measure a different variant with its own test and the figure runs several times higher. The number is not wrong. It is just much smaller than the question it gets used to answer.

Conclusion

A man's alphabet came from a plastic toy on a fridge door. Eleven more had the same story. Four hundred out of six and a half thousand did, and none among people born before the toy existed.

Meanwhile forty-three families carry the trait through generations, three more were sequenced down to thirty-seven candidate genes, and eighteen synesthetes in a scanner showed more coherent white matter than their matched controls.

Both are true, and the field has not finished reconciling them. What it has finished is testing the memory claim, and the answer is a small, specific, durable advantage that behaves like a second code and not like a better brain.

Which is the useful part. Not that some people get an extra handle on their memories for free, but that the handle is a mechanism rather than a gift, and the mechanism is one you already have access to. A second code, attached to the same material, giving retrieval somewhere else to start from. Theirs arrives with the letter. Yours has to be built. Whether a built one ever matches an involuntary one is not a question these studies answered, and I would rather leave it open than close it for you.