The Von Restorff Effect: Distinctive Items Survive, Similar Ones Blur

Introduction

Write down eight things you need from the shop. Milk, bread, eggs, butter, rice, tea, flour, and a birthday candle. Come back an hour later and try the list from memory. You will very likely get the candle.

That is the effect this article is about, and you have met it a hundred times without a name for it. One red word on a page of black ones. The one question in an exam that was not like the others. Psychologists call it the von Restorff effect, or the isolation effect, and it is one of the most repeated findings in popular writing about memory.

The version that circulates goes like this. The odd item stands out, so it grabs your attention, so it gets remembered. Then a design tip. Make your button orange. Highlight the key sentence.

That story is not exactly wrong. It is also not what Hedwig von Restorff argued, and the difference matters more than it sounds. Her account leaned on the seven items you forgot rather than the one you kept. Items that resemble each other get in each other's way. The candle was never helped so much as it was spared. R. Reed Hunt found this reading so thoroughly lost that in 1995 he published a paper about it and called it The subtlety of distinctiveness: What von Restorff really did [1].

The distinction is not academic housekeeping. If the odd item wins partly because the others lose, distinctiveness is not a bonus you can hand out freely. It is a budget. Marking one sentence in a page is a memory instrument. Marking half the page is a colouring activity, and the reason is mechanical rather than moral.

There is more. In 2026 a team had to rescue this effect from its own literature, because it had been quietly failing in recognition tests for years in a way that fitted no theory anybody had [2]. And in one careful set of experiments the odd item did not merely fail to win. It lost [3].

This article sets out what ninety years of research actually found. Where the popular version is accurate, it says so. Where the field is still arguing, and it is arguing in at least six places, it names who is on which side.

What Hedwig von Restorff Actually Did

Start with the paper, because almost nobody quotes it.

Hedwig von Restorff was a physician and psychologist working in Berlin in the early 1930s. Her famous study was not a study in the modern sense. It was her doctoral dissertation, written under Wolfgang Köhler, and it carried the title "Über die Wirkung von Bereichsbildungen im Spurenfeld" [4]. Roughly: on the effect of the formation of a field of items in the trace field. Köhler co-authored the follow-up analysis four years later [5].

That title is worth sitting with, because it tells you what kind of research this was. Köhler was one of the founders of Gestalt psychology, and the Gestalt programme was about structure. Not about which item is brightest. About what happens to a whole when you rearrange its parts. The Spurenfeld, the trace field, was their name for the organised set of traces a list leaves behind. Von Restorff was asking what a block of similar items does to that field.

She was not asking how to make a shopping list stick. She was not writing about advertising. Nothing in the original is a tip.

Colin MacLeod's 2020 history of the two most famous memory effects named after women, hers and Bluma Zeigarnik's, makes the same point about both [6]. Each has become a slogan. Each had a specific argument underneath it that the slogan drops.

A slogan is easier to carry than an argument, and it travels a great deal further.

The dissertation has never been published in full English translation. For most of the ninety years since, English-speaking psychology has known von Restorff through summaries of summaries, and the summaries kept the finding and threw away the explanation.

By 1965 the phenomenon had enough literature behind it that W. P. Wallace reviewed its historical, empirical and theoretical status in Psychological Bulletin and concluded that no theory on offer settled it [7]. That was sixty years ago. The theories still do not settle it, which is worth knowing before anybody calls this a law.

The Explanation That Got Lost

Here is the sentence that most pages on this topic never print.

Von Restorff's account of the isolate advantage put more weight on what happened to the non-isolates than on what happened to the isolate. Seven pieces of furniture in a list interfere with one another. They share features, they share a category, they compete at retrieval. The chipmunk sitting among them shares nothing with anybody, so it never joins the scrum. It is not that the chipmunk got a boost. It is that the furniture got crowded.

Hunt's 1995 paper is the standard modern statement of this reading [1]. He had already argued in 1993, with Mark McDaniel, that organisation and distinctiveness are not opposites but two descriptions of the same underlying processing, which is a harder idea than either slogan [8].

Hunt and C. A. Lamb sharpened the test in 2001 by insisting on a proper control condition, and showed that the isolate beats both the items surrounding it and that control [9]. That design is what makes the question answerable at all, because without a control you cannot tell a rise from a fall.

You can see why the simpler story won. A spotlight is easy to picture. Interference among seven items you were not paying attention to is not.

But the two accounts make different predictions, and people have been testing them for seventy years. R. T. Green asked in 1956 whether plain surprise was doing the work, and came back with an answer more complicated than the question [10].

Surprise sounds like the obvious answer until somebody tries to measure it. It dissolves into whatever else the task is already doing.

Eli Saltz and Slater Newman tested the intralist association assumption directly in 1959, which is the interference side of the argument stated formally [11]. Arthur Jensen stripped the task down to minimal response learning in 1962 to see what the effect actually needs in order to survive [12].

None of these people were arguing about a colour. They were arguing about what a list does to itself.

Darryl Bruce and Marion Gaines put an organisational hypothesis on trial in 1976, asking whether the isolate wins because it cannot be grouped with anything [13].

Notice the shape of the question. It keeps returning to the group, not the winner.

None of that settled it either.

The modern version of the argument is more direct and very recent. In 2026 Sébastien Gionet and colleagues ran 240 participants through four experiments with 10-word lists, immediate and delayed, to test distinctiveness against interference as explanations of the same free-recall advantage [14]. Jean Saint-Aubin's group had already built a model of the short-term version and given it a name that says the whole thing in four words. They called it the cost of relative distinctiveness [15], and extended the same model to backward recall in 2026 [16].

Cost. Relative. Two words the design blogs never use.

The left branch is von Restorff's own account. The right one is the distinctive-processing reading that displaced it in the textbooks. They agree about the item you remember and disagree about the items you do not, which is why the argument has lasted ninety years. You cannot tell them apart by looking at the winner.

Two Ways to Be Different

The word distinctive is doing a lot of work here, and it covers at least two different things.

Stephen Schmidt drew the line in 1991 in a paper with a title that is still a fair summary of the field's problem: can we have a distinctive theory of memory [17]. Primary distinctiveness is difference against the immediate background. One red word among black ones. Secondary distinctiveness is difference against everything you know. A cow on a skateboard is odd whether or not anything around it is.

Those two kinds come apart the moment you try to build a study technique on either one of them.

They do not behave the same way, and the second kind is a good deal messier. Mark McDaniel and colleagues spent a 1995 paper on the bizarreness effect and settled on a title that is also a verdict: it is not surprising, it is complex [18].

Oddness is not a property that an item can carry around on its own. It is a relationship, and the relationship can be with the page or with the world.

Then there is the finding that quietly kills the simplest version of the popular story. John Dunlosky, R. Reed Hunt and Elizabeth Clark asked in 2000 whether perceptual salience is needed to explain the isolation effect at all [19]. On their evidence it is not. A conceptual isolate with no visual difference whatsoever, a chipmunk printed in the same font and the same colour as the furniture, produces the effect. Nothing is brighter. Nothing is bigger. The item is simply about something else.

Read that again if you have ever been told this effect is really about visual contrast.

So the orange button is one instance of a broader thing, not the defining one. Hunt made distinctive processing the centre of his 2013 statement of the position, framing it as precision rather than prominence [20], and his opening chapter for the 2006 volume on distinctiveness had already spent its length on how slippery the concept is [21]. James Van Overschelde and colleagues showed in 2005 that distinctive processing is what memory experts are doing when they perform feats that look like raw capacity [22].

Precision is a better word than prominence, because it points at what the system does with the item rather than how the item looks.

All of this sits inside an older frame. Fergus Craik and Robert Lockhart's levels-of-processing paper from 1972 is the ancestor of the argument, and whether depth and distinctiveness are the same thing under two names has never been settled [23].

That last row is not a technicality. It is the boundary condition that decides whether any of this transfers to your own reading.

Where It Sits in the List

Ask most people why the odd item wins and you get an answer about attention. Push a little and you get a second: the odd item gets more rehearsal. You notice it, so you dwell on it, so you say it to yourself a few more times.

That answer is testable and it does badly.

Von Restorff put her isolate early, at around the second or third position. That already makes the rehearsal story awkward, because early items get rehearsed most anyway. Ronald Bone and L. R. Goulet went at the question directly in 1968 by moving the isolate along the list and watching what happened to the advantage [24]. The serial position curve itself, the U shape that gives you good memory for the start and end of a list and poor memory for the middle, was characterised by Bennet Murdock in 1962 [25]. The serial position effect is worth reading on its own terms, because the two phenomena share a list and not much else.

The list is not a queue where the loudest item gets served first.

Lisa Geraci and colleagues went further in 2010. They gave participants delayed judgements of learning, which is a fancy way of asking people to predict what they will remember, and found recall better for the isolated item regardless of where it sat in the list [26]. The predictions tracked the effect too, for both early and late isolates. People could feel it coming.

That last part matters more than it looks. People are not blind to their own encoding, which cuts against the idea that this all happens beneath awareness.

Then Stephen and Constance Schmidt complicated it. Across three experiments in 2017 they found that under moderately distracting conditions the isolation effect appeared for midlist items and not for early ones [3]. That much is a boundary. The next part is stranger. Early conceptual isolation actually impaired memory for the distinct item, relative to the same item in a list where everything matched. A second experiment reproduced the impairment with no distraction at all.

Read that again. The odd item, placed early, was remembered worse than if it had not been odd.

Those two results are in direct conflict. Geraci found the advantage regardless of position, Schmidt and Schmidt found an early conceptual isolate impaired, and nobody has settled which condition explains the gap.

Their third experiment found the split. A number isolated in a list of words survived early placement, and so did a word among numbers, because that contrast is coded automatically. A conceptual contrast, chipmunk among furniture, needs relational processing and does not survive being first. Asking people to make judgements of learning shrank the early effect further.

The picture is not a spotlight. It is a system comparing an item with a context that has not fully arrived.

It Depends What You Are Asked to Do

Now the part that no page-one result mentions, and the reason this article exists in 2026 rather than 2016.

For decades the isolation effect was reliable in free recall and hard to find in recognition. Gerrit van Dam and colleagues had reported the split as early as 1974 [27]. Free recall means you are handed a blank page. Recognition means you are handed the words and asked which ones you saw. If distinctiveness makes an item easier to find, the split makes sense, because recognition hands you the item and there is nothing left to find. If distinctiveness strengthens the trace, recognition should show it too.

Siri-Maria Kamp and colleagues put the problem bluntly in 2026. As they write, there is rarely an isolation effect for recognition, and that absence "is inconsistent with intuition and all known theories of memory" [2].

That is a remarkable sentence about a phenomenon most websites describe as settled.

Their solution was not a new theory. It was better experiments. Three studies extended the standard design to a multilist procedure and confirmed the free recall advantage for unexpected words, with no recognition advantage unless recall happened first. Source memory gave an ambiguous answer. Three further experiments then collected many more observations per person in the critical condition and controlled the order in which the tests were given, and under those conditions the von Restorff effect appeared in recognition and in source memory after all.

Six experiments to establish that a ninety-year-old effect exists in a second kind of test. Recognition and recall diverge more often than people expect, and this is a clean example.

There is a counterweight from the same year. Nathan Gillespie and Gregory Cox ran four experiments on artificial auditory textures and found recognition affected by similarity, by serial position and by list homogeneity, but not by how distinctive people rated the individual sounds [28]. The structure of the list mattered. The subjective oddness of the item did not.

What the Brain Does With the Odd One Out

The electrical signature came first, and it came early.

In 1984 Demetrios Karis, Monica Fabiani and Emanuel Donchin recorded brain activity while people studied lists containing isolates, and found that the isolated item produced a larger P300, a positive wave arriving roughly a third of a second after the stimulus [29]. Two years later Fabiani's group showed that P300 amplitude at study predicted whether a word would later be recalled, in a task where people did not know their memory would be tested [30]. By 1995 Fabiani and Donchin had built the observation into a model of the von Restorff effect itself [31].

Take the qualifier seriously. The relationship is not clean. Leun Otten and colleagues examined the link between P300 amplitude and later recall for distinctive events in 2000 and found it depended on the task rather than holding across the board [32].

Correlations tell you something is happening. They cannot tell you the brain area is doing the work. For that you need a case where a piece is missing.

That is what Kishiyama, Yonelinas and Robert Knight did in 2009. They tested 16 patients with damage to lateral prefrontal cortex on a von Restorff novelty paradigm [33]. The patients were impaired on both recollection and familiarity, and, crucially, they did not show the normal memory advantage for novel items at all. The effect was gone. Kishiyama and Yonelinas had already shown in 2003 that novelty affects recollection and familiarity differently in healthy people [34].

Sixteen people is a small study. It is also the closest thing to causal evidence this literature has, because the alternative designs cannot separate correlation from cause, and that trade-off is worth saying out loud.

Around the same effect sits a much larger literature on novelty and the medial temporal lobe. Endel Tulving and Neal Kroll proposed in 1995 that novelty assessment works as a gate on what gets encoded at all [35], and Robert Knight showed in 1996 that damage to the hippocampal region reduces the electrophysiological response to novel events [36].

Notice that the question has quietly widened, from lists of words to events in a life.

Imaging filled in the map. Reza Habib and colleagues separated the novelty and familiarity discrimination networks in 2003 [37], the same year Charan Ranganath and Gregor Rainer reviewed the neural mechanisms for detecting and remembering novel events [38].

Mapping a network is not the same as showing that it does the job.

Then the circuitry. John Lisman and Anthony Grace described a hippocampal to midbrain loop in 2005 that would let novelty control what enters long-term memory [39], and a 2016 systematic review pulled together the neurotransmitter evidence behind that loop [40].

This is the machinery that would let a surprise buy its own way into storage.

The animal work is worth flagging as animal work. A 2020 mouse study identified a hypothalamic novelty signal that modulates hippocampal memory [41], and a 2022 paper found a synaptic signal for novelty processing in the mouse hippocampus [42]. Those are mechanisms in mice. They are not a human count.

The gap between a mouse hippocampus and yours is where a lot of confident writing goes wrong.

Human imaging has got much more precise. In 2025 Jörn Quent and colleagues scanned 56 participants at 7 tesla while they moved through a virtual arena and found that the brain encodes spatial novelty in graded form rather than as a switch between new and old [43]. A 2026 meta-analysis pooled 66 experiments and 1771 subjects and separated two modes of novelty processing, an active one that recruits goal-directed control and a passive one driven by bottom-up reorienting [44], and a 2026 imaging study found a two-level hierarchy in how the brain handles auditory novelty [45]. Even the word novelty names more than one process, and the newer imaging says the brain treats novelty as a matter of degree rather than as a switch.

The through line is that noticing is not optional. Much of what follows depends on what attention does at the moment of encoding, and the prefrontal patients are the sharpest evidence for it.

That is not the folk story creeping back in. Attention is a gate the isolate has to pass through. It is not what produces the advantage on the other side.

Not Everyone, and Not Equally

The effect shows up in children. Richard Cimbalo's group demonstrated it in children's short-term memory in 1981 [46].

Which is what you would expect if this runs on ordinary encoding rather than a skill that has to be learnt.

Older adults were a puzzle for twenty-five years. In 1982 Cimbalo and Lois Brink looked for the isolation effect in older adults and did not find it [47]. That null result sat in the literature and got cited.

A null result is not nothing. It sets the terms for whoever comes next, and occasionally it sets them wrongly.

Then in 2008 Tamra Bireta, Aimée Surprenant and Ian Neath ran the comparison again using a physical manipulation, font colour, and found a significant von Restorff effect in both younger and older adults [48]. The effect size was smaller in the older group. That is not a failure of the earlier study so much as a difference in method, and the smaller size fits a specific prediction. Moshe Naveh-Benjamin's associative deficit hypothesis, published in 2000, holds that ageing reduces the ability to bind an item to its context [49]. An isolate is defined by its relationship to a context. If binding weakens, the advantage should shrink rather than vanish. It shrank.

Shrinking is a far more informative result than vanishing, because it tells you which part of the process is fading and which part is holding.

One caveat travels with it. Bireta's group used font colour, a perceptual contrast, and perceptual and conceptual isolates come apart. Whether older adults hold conceptual isolation as well is a separate question.

More recent imaging fits the same shape. In 2023 Claire Pauley and colleagues scanned 39 younger adults aged from 18 to 27 years and 37 older adults aged from 64 to 76 years during incidental learning and a surprise recognition test, and found age differences in how distinct the neural patterns were at encoding and retrieval [50]. The representations themselves are less separated. There is more on how memory changes with age if you want the wider picture.

Less separated representations mean fewer things that can stand out from one another. The background stops being uniform.

The effect survives in clinical populations too. A 2019 study of the Brazilian Rey Auditory Verbal Learning Test slipped one emotionally weighted word, the word for mother, into an otherwise neutral list, and found the von Restorff effect in three groups: n = 114 healthy controls, n = 87 with mild cognitive impairment and n = 86 with Alzheimer's disease [51]. The serial position curve in all three groups came out W-shaped rather than the classical U, because the isolate creates a bump in the middle. A related 2022 study had 499 participants across four diagnostic groups complete a novelty fMRI task, with responses in precuneus and medial temporal regions differing by risk status [52]. These are findings about groups, not something you can apply to yourself.

The Bill

Now the part that changes what you do on a Tuesday evening with a textbook.

If part of the isolate's advantage is the crowd's disadvantage, distinctiveness has to be paid for out of something. The clearest demonstration that salience carries a price comes from a different paradigm altogether. It is an emotion study.

In 2003 Strange, Hurlemann and Dolan showed people lists of words containing an occasional emotional one, and found that memory for the emotional word was enhanced while memory for the words immediately preceding it was impaired [53]. Both effects were reversed by propranolol, a beta-blocker, which points to a shared noradrenergic mechanism. The salient item did not just win. It took something. A 2024 follow-up with a required sample of n = 70 participants examined how emotional salience shapes the forward flow of memory across a sequence [54]. Whether that trade-off is the same one at work in an ordinary isolation list is not established. How emotion shapes what you keep is a close cousin of everything here.

Which brings us to the highlighter.

John Dunlosky and four colleagues reviewed 10 common study techniques in 2013 for Psychological Science in the Public Interest, and only 2 of the 10 came out with a high utility rating [55]. Those two were practice testing and distributed practice. Highlighting and rereading were included on purpose, precisely because students report relying on them so heavily, and both landed among the 5 of the 10 techniques rated low utility [55]. The experimental ancestor of that verdict is Robert Fowler and Anne Barker's 1974 study of highlighting and retention of text material [56].

The mechanism is now easy to state. A highlighter does not add memory. It redistributes it. Mark one sentence in a page and you have created an isolate. Mark forty and you have two uniform groups and no isolate anywhere. The relative in relative distinctiveness is doing most of the work here.

Making something merely different does not work if the difference carries no meaning. In 2025 Brady Roberts and colleagues tested memory for chemical elements across six experiment groups with 821 participants in the final analyses, comparing element names, chemical symbols and meaningless letter strings [57]. The design was powered to catch an effect as small as d = 0.47, so a real symbol advantage would have been hard to miss [57]. Among students who had passed an introductory chemistry course, memory for words and for symbols was equivalent, and both beat meaningless letters. Symbolism by itself bought nothing. What the symbol meant did.

Meaning is doing the work, and the surface is where you notice it.

The version that does work is the one that changes the structure of what surrounds the material. In 2022 Joey Essoe and colleagues had 48 participants learn 80 foreign words in two phonetically similar languages inside desktop virtual reality, half of the participants learning both languages in one environment and half learning each language in its own [58]. Among the subgroup who reported feeling genuinely present in the environment, the two-context group retained 92 percent of the words after a week against 76 percent for the one-context group [58]. That is a gap of 16 points, and the manipulation behind it was nothing more than making two similar bodies of material stop resembling each other [58]. Read it carefully, though: those numbers come from that subgroup and not from the sample as a whole, so this is interference between similar memories being cut at the source under favourable conditions, not a guarantee.

Nothing was highlighted in that study. The change was to the room.

Related work points the same way. The production effect, where reading words aloud improves memory for those words, runs on the same distinctiveness logic and carries the same accounting problem, both in recognition [59] and under instructions to forget [60].

Saying one word aloud makes it odd among words you only read. Say every word aloud and the oddness is gone again.

Daniele Saraulli and colleagues put the question to a clean test in 2023. The attentional boost effect, where words paired with a target you must respond to are recognised better, was significant when both kinds of word sat in the same list and disappeared when the two kinds were split between separate groups [61]. Take away the contrast and you take away the effect. A 2024 study found the same logic behind the advantage of pictures over environmental sounds, with correct responses higher for pictures across targets and foils alike [62].

The same accounting keeps surfacing in corners of the field that did not set out to test it.

Lorents, Ruitenberg and Schomaker reviewed in 2023 how novelty can boost memory for material met before or after the novel event, and said plainly that the human evidence still lags the animal evidence [63]. A 2026 study followed prior novelty into later target detection [64], and another tested how novelty and rest act on either side of learning [65].

The animal caveat is worth holding on to. It separates a promising mechanism from something you can act on tonight.

Melanie Sekeres and colleagues framed the underlying choice neatly in 2024. Whether the brain updates an existing memory or builds a new one depends on how far the event departs from what you already know [66].

The practical shape of all this is not a technique. It is a constraint. You cannot make everything important. Deciding what to mark is deciding what to lose, and the cues you build at encoding are what you will have to work with later.

When It Fails

Several things reliably break the effect, and none of them is obscure.

Noticing comes first, and it is not automatic. Schmidt and Schmidt tested this in 2015 by putting isolates behind an easy or a difficult Stroop task [67]. As they open the paper, "sometimes we fail to notice distinctive or unusual items (inattentional blindness), while other times we remember distinctive items more than expected items (the von Restorff effect)". Highly arousing taboo words survived even the hard version. A conceptual isolation effect appeared only when the category change was made relevant by the task, or the task was easy, or the isolated items had a retrieval advantage anyway. Distinctiveness that nobody registers is not distinctiveness.

An isolate you never registered is just a word.

Position breaks it, as the 2017 experiments showed. So does the kind of contrast: in those experiments automatic ones survived early placement and conceptual ones needed context that had not yet accumulated [3].

So the effect needs a context, and a context takes time to assemble.

Expectation complicates it, and that has become its own research programme. A 2023 study reframed prediction error as a continuum rather than a switch, running from events that match what you expected through to events that flatly violate it [68], and a 2026 experiment found expectation congruence pushing recall and recognition in opposite directions [69].

A continuum is what makes the practical advice hard. There is no threshold you can aim for.

Darya Frank and Alex Kafkas showed in 2021 that novelty effects on episodic memory depend on what you were expecting to see [70], and Kafkas found in the same year that pupil size at encoding separates expected from unexpected novelty [71]. A 2026 study added emotion and list context as further modulators [72].

Your pupils are keeping score before you have any idea that they are.

The shape that keeps appearing is a U rather than a line. In 2025 Gerrit Höltje and colleagues tested N = 36 participants on sentence endings and found that both strongly expected and highly unexpected words were recognised better, while moderately expected words were remembered worst [73]. Those participants separated studied words from new ones cleanly, at d = 2.14, so the sagging middle is not a measurement problem [73].

Being slightly unusual is the worst place to be.

The comfortable middle is where memory goes quiet.

A 2025 registered report ran the same idea backwards. Using a surprise memory test disguised as a word-judgement task, it found that words about the body produced more false alarms even when nobody was attending to them, which the authors read as those words being less distinctive as a memory trace rather than more [74]. Distinctiveness can be lost as well as gained.

And a 2026 study of 48 participants split into two groups of 24 showed that reward associations alone can make abstract visual stimuli memorable with no distinctiveness manipulation at all, learning held well above the 33 percent chance level [75]. Memorability has more than one supplier.

And distinctiveness can misfire in the other direction, by making you confident. Daniel Schacter's group described a distinctiveness heuristic in 1999, where people use the absence of vivid detail as evidence that an event did not happen [76], and David McCabe and colleagues followed it into false recall in 2006 [77]. It usually helps. When it fails it fails confidently, which is worse. Tim Valentine's 1991 face-space account explains the same asymmetry for faces, where distinctive faces are recognised better and typical ones are confused with each other [78]. The eyewitness implications are real and belong in a different article.

Confidence and accuracy come apart quietly here, which makes this failure dangerous rather than merely wrong.

One more, from a different direction. Maverick Smith and colleagues found in 2024 that directing attention to the boundaries between events improves memory, and that the benefit differs across the lifespan [79]. Structure helps. Structure is also what an isolate disrupts. The field has not fully reconciled those two facts.

Two true findings pulling in opposite directions is not a scandal. It is where the next round of experiments comes from.

What Ninety Years Adds Up To

Here is what you can say without hedging.

Put one item in the middle of a list that shares nothing with the others, ask people to recall the list, and the odd item comes back more often. That has held since 1933 across colour, category and stimulus class. What happens at the start of a list is a separate question and it is still open.

It is one of the sturdier findings in memory research, which is why the loose talk around it is worth correcting. The odd item produces a larger P300 at study. Lateral prefrontal cortex is doing part of the work, on the evidence of 16 patients in whom the advantage is absent. Children show it, on one 1981 study. Older adults show it too, at least for a perceptual contrast. You have to notice the item for any of it to happen.

Here is what is still open, and it is more than the popular version admits. Whether the advantage is mostly a gain for the isolate or mostly a loss for its neighbours. Whether distinctive processing or better retrieval cues explains it, which Qian Wen Chee and Winston Goh set up as a direct contest in 2018 and did not resolve [80]. Whether the 2026 recognition result holds up, given that it overturns decades of null findings on the strength of a design change. Whether an early isolate helps or hurts. Whether the smaller effect in older adults really comes from weaker binding or from something else. Whether the boost spreads to neighbouring material in humans the way it does in animals.

Six open questions is not a weakness in the science. It is what a real phenomenon looks like from the inside, and it is why a page calling this a law of design is telling you less than it appears to.

A law does not need six open questions. A living research problem does.

What survives all of it is the thing the title says. The odd item survives because the others blur, and the blur is not a bug in your memory. It is what a memory system does when you hand it eight things that are almost the same. You cannot switch that off. You can only choose, quite deliberately and quite sparingly, what you are going to make different.