Epilepsy and Memory: The Surgery That Cost H.M. Every New Memory

Introduction

In August 1953 a 27 year old assembly line worker from Connecticut lay on an operating table in Hartford Hospital while a surgeon opened two holes above his eyes and removed a piece of brain from each side of his head.

His name was Henry Molaison. He had been having seizures since he was ten, and by his late twenties they had taken his job, his independence and most of his life. Drugs had failed. What was left was an operation nobody had performed for this reason before.

It worked. The seizures got dramatically better.

He also never formed another new memory, for fifty five more years.

Almost everything the twentieth century learned about how human memory is organised came out of that trade. The distinction between the memories you can describe and the skills you cannot. The idea that a specific patch of tissue does a specific job. The reason a person can be perfectly intelligent and still be unable to tell you what they had for breakfast. All of it traces back to one man and one afternoon.

Here is what almost nobody tells you. It was an epilepsy operation. Not a psychiatric one, not an experiment on a healthy brain, not a lobotomy. A neurosurgeon was trying to stop seizures, and the memory loss was collateral damage that nobody predicted.

That fact has been quietly cut out of both halves of the internet. Search for epilepsy and memory and you get patient information pages from hospitals and charities that never mention Henry Molaison once. Search for patient H.M. and you get psychology explainers that stop somewhere around 1968 and never say a word about what epilepsy surgery does to memory today. The two stories belong to each other. Nobody has joined them.

This article joins them. It runs from a 1953 operating theatre to a 2014 laboratory where somebody finally cut that brain into 2,401 slices and discovered the textbooks had the anatomy wrong, and then into a clinic in 2026 where a real and treatable memory disorder is being missed because the standard test is run at the wrong interval.

What Epilepsy Actually Is, and Why It Reaches for Memory

A seizure is a burst of abnormally synchronised electrical activity in the brain. Epilepsy is the tendency to have them repeatedly without an immediate provocation. Beyond that the word covers an enormous range of conditions, which is why the International League Against Epilepsy has revised its classification several times, most recently in 2025 [1][2].

Tens of millions of people worldwide live with it [3][4]. For most of them, medication works. For a substantial minority it does not, and that minority has a formal name. Since 2010 the field has defined drug resistant epilepsy as the failure of two appropriately chosen and properly tolerated medication schedules to achieve sustained seizure freedom [5]. Two failures. After that the odds of a third drug working drop sharply, and the conversation changes.

Now the part that matters for this article. The single most common form of drug resistant epilepsy in adults starts in the temporal lobe, and specifically in its inner edge. That inner edge is called the medial temporal lobe, and it happens to contain the hippocampus, the structure that decides what gets remembered.

So the tissue that generates the seizures and the tissue that builds new memories are the same tissue. That is the whole problem in one sentence.

Everything else in this article follows from that overlap. If the seizures came from the back of the head you would be reading about vision. They come from the memory system, so you are reading about memory.

It is also why temporal lobe seizures feel so strange from the inside. Patients report intense familiarity, a sense of having lived this exact moment before, sudden vivid fragments of scenes from years ago. Pierre Gloor and colleagues studied these experiential phenomena directly with electrodes in the limbic system in 1982 and found they arise from the same structures that support ordinary recollection [6]. A seizure in the memory system produces a symptom made of memory. The clinical picture was mapped in detail through the early nineties [7].

Henry Molaison's seizures came from exactly there. Both sides.

The Man Who Was Not the Only Patient

Popular retellings present H.M. as a lone case, a singular experiment. He was not.

Scoville had been performing medial temporal resections for several years before 1953, mostly on psychiatric patients. When he and Brenda Milner published the paper that made the case famous in 1957, Henry Molaison appeared as one patient inside a series of ten [8]. The other nine had been operated on for psychosis. That detail rarely survives the retelling, and it should, because it changes what kind of story this is.

Scoville was not conducting a memory experiment. He was doing what surgeons of that era did with a technique they believed in, on patients who had run out of options and, in most cases, could not meaningfully refuse.

What made Molaison different was that he was cognitively intact going in. His intelligence was normal. His personality was unremarkable. When the memory deficit appeared afterwards, there was nothing else to blame it on. In the psychotic patients the same deficit had been there and had been invisible, buried under everything else that was wrong.

The cause of his epilepsy was never established. You will read almost everywhere that a childhood bicycle accident did it. The primary literature does not support that. A head injury appears in his history. Whether it caused the seizures was never determined and probably cannot be.

He was 27 years old. He agreed to the operation. Nobody involved, including him, had any idea what he was agreeing to.

It is tempting to read that as villainy. It was not. It was a surgeon working at the edge of what anyone knew, on a patient whose life had been taken over by seizures, with no imaging and no precedent to warn him. The mistake was real. So was the reason for making it.

What Scoville Removed

The operation removed tissue from the medial temporal lobe on both sides: parts of the amygdala, the entorhinal cortex, and a length of the hippocampus.

Scoville estimated the resection extended about eight centimetres back from the temporal pole. That figure has been repeated for seventy years as though someone measured it. Nobody did. It was a surgeon's estimate written down after an operation, and when Suzanne Corkin and colleagues finally imaged Molaison's brain with MRI in 1997, the lesion looked closer to five centimetres, with a meaningful amount of posterior hippocampus still present [9].

Hold on to that. It becomes important later.

There is a general lesson buried in it too. A number gets written down once, by someone with no reason to be precise, and then it gets copied for seventy years by people who assume somebody checked. You will find that pattern in most fields if you look for it.

The immediate result was what the 1957 paper describes with startling understatement. Molaison could no longer form new memories of events or facts. He could still hold a conversation. He could still tell you about his childhood. He simply could not add anything new to the record [8].

One year later Wilder Penfield and Brenda Milner reported two more patients who had developed the same profound amnesia after resection on only one side. Both turned out to have unsuspected damage on the other side already [10]. That established the rule that stopped this from ever happening again by accident. One side is survivable. Both sides are not.

The Year the Word Memory Changed Meaning

Before 1953 memory was treated as a general property of the brain, spread everywhere, a bit like intelligence. Karl Lashley had spent decades cutting bits out of rat cortex looking for the place where a learned maze was stored and had concluded it was not anywhere in particular.

Molaison ended that argument in one case.

You do not usually get to end a decades old scientific argument with a single person. It happened here because the deficit was so specific that no general theory could absorb it.

Here was a man whose intelligence was intact, whose language was intact, whose personality was intact, whose memories from before the operation were largely intact, and who could not learn that his uncle had died. Memory was not a general property. It was a job, and a specific piece of tissue did it.

That is the finding that built cognitive neuroscience. Larry Squire and Stuart Zola spent the following decades turning it into a system with a name and a map [11][12]. In 1986 they reported patient R.B., a single case with damage restricted to one subfield of the hippocampus called CA1, who had a real amnesic syndrome [13]. One subfield. That is how precise the anatomy turned out to be.

Note the sample sizes. R.B. was one patient. Molaison was one patient. Some of the most consequential findings in the history of neuroscience rest on n equal to one, and they hold up because the deficits were so specific and so reproducible across decades of testing, not because the numbers were large.

Look at the gap between 1968 and 1997. Nearly thirty years in which the field taught the case confidently without anyone having seen the actual lesion.

The Mirror and the Star

In 1962 Brenda Milner sat Molaison down with a five pointed star drawn on paper, a pencil, and a mirror. His task was to trace between the two outlines of the star while seeing only the mirror image of his own hand. It is genuinely difficult. Everyone is terrible at it to begin with.

He did it for three days.

On day two he was better than day one. On day three he was better than day two. His improvement curve looked like anybody else's. And every single morning he greeted the task as something he had never seen before [14].

Stop and consider what that means. His hands had learned something his mind had no access to. The improvement was real, measurable and permanent. The memory of acquiring it did not exist.

You have a version of this yourself, without the amnesia. Try to describe exactly what your hands do when you tie a shoelace. The knowledge is clearly in you. It is not in the part of you that talks.

That result split memory in half and the halves have never been rejoined. On one side sits everything you can declare: events, facts, the name of the street you grew up on. On the other sits everything you can only demonstrate: riding a bicycle, typing, tracing a star in a mirror. The cerebellum and basal ganglia carry the second kind, which is why Molaison kept it. His surgeon never went near them.

Marvin Chun and Elizabeth Phelps later showed the same dissociation in a cleaner form. Amnesic patients with hippocampal damage learn implicit contextual regularities in a visual search task perfectly well while having no idea any regularity exists [15]. The learning happens. The knowing does not.

If you want the fuller version of what survives when explicit recall fails, we have written separately about the memories you cannot recall but still act on.

What He Kept

The line that circulates is that Henry Molaison had no memory. That is wrong, and the ways it is wrong are the interesting part.

He could hold a number in mind and repeat it back. He could carry on a conversation, follow an argument, make a joke land. Working memory over seconds was untouched, which is why people meeting him briefly often noticed nothing at all.

He could tell you about the 1930s. His childhood, his parents, the world before the war. Remote memory from long before the operation was largely there, though it thinned as it approached 1953. That gradient is itself a finding, and it is the reason a whole theory of memory consolidation exists: memories appear to migrate away from the hippocampus over years [16][17][18].

He learned a few new facts, very slowly and very partially, over decades of repeated exposure. He knew something had happened to his memory. Suzanne Corkin, who studied him for close to fifty years, documented what he could and could not acquire in unusual detail [19].

So the honest description is not that he had no memory. It is that one specific manufacturing process had been switched off while everything already manufactured stayed on the shelves. You can talk to someone in that state for ten minutes and notice nothing.

And in 2007 Demis Hassabis and colleagues found something nobody had thought to test. Patients with hippocampal amnesia cannot imagine new experiences either [20]. Asked to describe a beach they have never visited, they produce fragments with no coherent scene. The structure that assembles the past turns out to be the same one that assembles the future. It is not a filing cabinet. It is a construction site.

Rows one and two did not behave identically, and neither did rows one and five. Those gaps are the reason episodic and semantic memory are treated as separate systems at all [21], and the reason declarative and procedural memory are treated as separate systems on top of that.

Then Somebody Looked, Slice by Slice

Henry Molaison died in December 2008. He had consented, in whatever sense a man with no ability to form new memories can consent, to the donation of his brain.

This is the moment the story turns, so it is worth being clear about what happened. For fifty five years the field had described a lesion nobody had seen. Everything was inference from a surgeon's notes and, later, from scans of a living brain that could only show so much.

His brain was frozen, then cut into 2,401 histological slices, then photographed and rebuilt as a digital three dimensional model. Jacopo Annese and colleagues published the result in 2014, and it did not say what everyone expected [22].

A substantial amount of posterior hippocampus had survived the 1953 operation. More than the MRI work had suggested, and in better condition, with recognisable cellular architecture still intact.

What had not survived was the entorhinal cortex. That was almost entirely gone.

There was also diffuse damage in the deep white matter, and a small circumscribed lesion in the left orbitofrontal cortex that had nothing to do with Scoville at all.

You have to sit with what that does to the textbook version. For fifty years students were taught that H.M. lost his hippocampus and therefore lost new memory. He did not lose his hippocampus. He lost the tissue that feeds it.

The entorhinal cortex is the gateway. Nearly everything the cortex wants the hippocampus to know arrives through it, and nearly everything the hippocampus computes leaves the same way. Remove the gateway and you have a functioning structure that cannot receive or send. On this reading the lesson of the case is about disconnection, not removal.

It is a reinterpretation and not a settled fact, and it should be read that way. But it is not idle. Animal work had already shown that layer three of the medial entorhinal cortex is selectively vulnerable in rat models of temporal lobe epilepsy, dying preferentially while neighbouring layers survive [23]. That is rat data and it does not transfer automatically to people. It does suggest that entorhinal loss is a recurring theme in this disease rather than a one off surgical accident.

Think about what this does to a sentence you have probably read somewhere. "The hippocampus is where memories are made." It is not exactly false. It is the kind of true that stops being useful the moment you look closely, because a structure that cannot receive anything is functionally absent whether or not it is physically there.

There is a third position worth naming, because two sided arguments in neuroscience usually turn out to have a third side. In 2026 Georgios Argyropoulos and colleagues argued that diencephalic integrity explains aspects of what has been called hippocampal amnesia [24]. That moves part of the explanation out of the temporal lobe entirely, into structures around the thalamus.

The Argument That Never Closed

The standard account, associated above all with Larry Squire, treats the medial temporal lobe as a single memory system with the hippocampus at the centre [25][26]. It is coherent, it explains a great deal, and it has been the majority view for thirty years.

It has a problem, and the problem is a patient.

In 2007 Ben Bowles and colleagues reported someone who had undergone anterior temporal resection that spared the hippocampus. This person showed impaired familiarity, the vague sense that something has been encountered before, alongside preserved recollection, the ability to retrieve the actual episode [27]. That is exactly backwards from what a single unified system predicts, and it is hard to explain away.

So: one system or two processes? Squire on one side. Bowles and the dual process camp on the other. The argument has not closed and this article will not close it.

What you should take from it is that the field built an enormous amount on one man's brain and is still checking the foundations. That is not a weakness. That is what checking looks like.

Anyone who tells you the memory systems question is settled is describing a textbook rather than a literature.

It helps to look at the mirror image case. In semantic dementia the anterior temporal lobe erodes and meaning goes first, while the ability to recall recent events holds on far longer than you would expect. Molaison lost the new and kept the meanings. Those patients lose the meanings and keep the new. Two opposite patterns from damage a few centimetres apart.

Meanwhile, in the Clinic

Here is where the two halves of the story meet, and where most writing about H.M. simply stops.

Epilepsy damages memory without any surgeon being involved.

The most useful single study on this followed 147 surgically treated and 102 medically treated patients with temporal lobe epilepsy from a baseline assessment to a follow up between two and ten years later [28]. Christoph Helmstaedter and colleagues found significant memory decline in 50 percent of the medically treated group and 60 percent of the surgical group, with non memory functions barely changing in either.

Read that again. Half the people who never had surgery declined.

That number reframes the whole surgical conversation. The comparison is not surgery against a stable baseline. It is surgery against a condition that is already taking something.

And then the part that gets left out of the scary version. Surgical patients who became seizure free recovered. Non memory function came back within a year, and memory function came back by the long follow up. Stopping the seizures did not only stop the seizures.

Why does uncontrolled epilepsy cost memory? There is not one answer, and anyone who gives you one is simplifying. The underlying pathology contributes, most often a scarred and shrunken hippocampus called hippocampal sclerosis [29]. Duration contributes, with longer histories associated with worse cognitive profiles [30]. Network reorganisation contributes, and modern imaging shows the disease is not confined to the hippocampus at all but reshapes cortical microstructure and connectivity across the brain [31][32][33].

There is also a finding that needs handling with care. Resected temporal lobe tissue from people with refractory epilepsy sometimes contains hyperphosphorylated tau, the protein associated with neurodegeneration, and the amount correlates with cognitive decline [34]. A 2026 systematic review found this is not an isolated observation [35].

This does not mean epilepsy causes dementia. It does not. Memory difficulty in epilepsy is common, has several causes at once, and is usually not progressive in the way a neurodegenerative disease is. What the tau work says is that the relationship between chronic seizures and long term brain health is more complicated than anybody assumed, and it is being studied. That is all it says.

The Spikes You Never Feel

This is the part that surprises people who have lived with epilepsy for years.

Between seizures, the epileptic brain does not go quiet. It produces brief abnormal electrical events called interictal epileptiform discharges. They last a fraction of a second. They produce no symptom you would notice. On an EEG they look like a sharp spike and then nothing.

They are not harmless.

The word interictal simply means between seizures. For decades it was treated as the safe part, the quiet stretch where nothing was going wrong. That assumption is the one that has come apart.

Jonathan Kleen and colleagues recorded directly from the hippocampus of patients while they performed a memory task. When a discharge happened during the moment of retrieval, the chance of a correct answer dropped [36]. The spike was invisible to the patient. The forgetting was not.

Three years later Jennifer Gelinas and colleagues showed a plausible mechanism in Nature Medicine. Interictal discharges induce coupling between the hippocampus and the cortex, hijacking the same communication channel that consolidation normally uses [37]. The pathological signal is not just noise on the line. It is using the line.

That distinction matters more than it sounds. If the spikes are noise, controlling them is cosmetic. If they are competing for the same machinery memory needs, then a person whose seizures are fully controlled but whose EEG is still busy is not actually in the clear.

The general principle had been argued for years by Gregory Holmes and Pierre-Pascal Lenck-Santini, who made the case that these abnormalities are a cause of cognitive impairment rather than a marker of it [38]. Which side of the brain they come from turns out to matter too [39].

And this may not be an epilepsy story only. In 2017 Alice Lam and colleagues put electrodes near the medial temporal lobe of Alzheimer's disease patients who had no history of seizures and found silent hippocampal seizures and spikes that scalp EEG had missed entirely [40]. What that means for memory loss in dementia is an open question. It is a very interesting open question.

None of this is visible to the person it is happening to. That is worth sitting with. Somebody can be well controlled, seizure free for months, and still be losing material to events they cannot feel.

The Drugs

Antiseizure medications have cognitive side effects. Some have very few. Some have a great deal. The differences between agents are real and well documented in the pharmacology literature [41], and large comparative trials have measured how the common drugs perform against each other on effectiveness and tolerability [42]. Prenatal exposure has its own separate and carefully studied literature [43].

That is as far as this article goes on medication, deliberately. Which drug suits which person is a clinical decision made with a neurologist who knows the seizure type, the other conditions and the whole history. Nothing written here should be read as guidance about anybody's own treatment.

What is worth saying plainly is that medication is one contributor among several, and it is the one patients most often blame for everything. Mood is another, and depression is common in temporal lobe epilepsy and independently degrades memory performance [44]. Sleep is another, and sleep quality in drug resistant temporal lobe epilepsy tracks memory function directly [45].

Five causes, tangled together, in one person. Untangling them for an individual is exactly what a neuropsychological assessment is for.

If you have ever been told that your memory problem is simply a side effect, that may be true and it may also be a third of the story.

The Operation That Kept Being Done

Here is the uncomfortable fact. After 1953, temporal lobe surgery did not stop. It expanded.

It expanded because it works. That sentence deserves a moment, because it sits oddly next to everything above it. The same class of operation that destroyed one man's memory has since given a great many people their lives back.

Here is the evidence for that claim. In 2001 Samuel Wiebe and colleagues ran the first randomised controlled trial, assigning 80 patients with temporal lobe epilepsy to surgery or to another year of medication. At one year, 58 percent of the surgical group were free of seizures that impaired awareness. In the medical group the figure was 8 percent [46].

A decade later the ERSET trial asked whether operating earlier is better. It randomised patients with mesial temporal lobe epilepsy who had failed only two drugs, rather than the usual twenty years of failure. Zero of the 23 patients kept on medication were free of disabling seizures in the second year. Eleven of the surgical patients were [47].

That result comes with a caveat that has to travel with it. Planned enrolment was 200 patients. The trial recruited 38 and was halted early for slow accrual. A small trial stopped early is suggestive, not definitive, and the authors said so themselves.

In 2017 a single centre randomised 116 children and adolescents with drug resistant epilepsy to surgery or continued medical therapy, and surgery was superior for seizure freedom at twelve months [48]. Long term follow up of operated children suggests surgery can change the cognitive trajectory rather than merely halting it [49]. Adult outcome data should not be applied to children, and the paediatric literature is deliberately kept separate.

And yet people wait. The typical referral for surgery still comes after roughly two decades of seizures. Expert consensus recommendations exist specifically to shorten that delay [50], which tells you how entrenched the problem is.

Part of the delay is fear, and the fear has a name. Everybody in this field knows what happened to Henry Molaison.

The Principle That Replaced H.M.'s Lesson

If you take one idea from this article into a conversation with a neurologist, take this one.

The modern question before temporal lobe surgery is not is this hippocampus diseased. It is can the rest of the brain cope without it.

Chelune named this in 1995 and called it hippocampal adequacy versus functional reserve [51]. The logic is counterintuitive and it is correct. If the tissue you are about to remove is badly scarred and doing almost nothing, removing it costs very little, because the other side has already taken over. If the tissue is relatively healthy and still working, removing it costs a great deal.

Which means the patients at greatest risk of memory decline are the ones whose memory works best going in.

Read that as a piece of practical logic rather than a paradox. You are not being asked what is broken. You are being asked what is still carrying weight, because that is what you lose if you cut it out.

That principle has held up for thirty years and it keeps being confirmed. Ojemann and Dodrill documented verbal memory deficits after left temporal lobectomy back in 1985 [52]. Trenerry and colleagues showed hippocampal volume on MRI predicts what happens to memory afterwards [53]. Silvia Bonelli and colleagues used functional MRI to predict the effect of resection before it happened [54]. The whole presurgical imaging enterprise exists to answer Chelune's question [55].

The prediction is still being sharpened. Preoperative functional MRI is being used to forecast memory decline after left sided surgery [56], and in 2025 support vector machines were trained to predict postoperative memory outcomes directly [57].

None of that machinery existed in 1953. Scoville had a clinical impression and a scalpel. If you want a single sentence for how far the field has moved, it is that the modern version of this operation spends more effort predicting the cost than performing the cut.

Which side is operated on matters enormously. Dominant side surgery, usually the left, puts verbal memory at risk. Non dominant side surgery affects visual and spatial memory less consistently, which Spiers demonstrated elegantly by testing how patients found their way around a virtual town [58]. Face and name association, a right hemisphere heavy task, shows its own regional pattern [59].

A recent refinement is that it may not be the grey matter that matters most. Work published in 2026 found that the white matter removed during resection predicts verbal memory deficits [60], which follows earlier findings that preoperative white matter network organisation forecasts decline [61] and that the risk profile differs between resection and ablation [62].

Which brings us to the obvious idea. If big operations cost memory, make the operation smaller.

Smaller Cuts, and Whether They Actually Help

Wieser and Yasargil proposed selective amygdalohippocampectomy in 1982: go in through a narrow corridor, take the medial structures, leave the lateral temporal lobe alone [63]. It sounds obviously better.

The evidence is genuinely mixed and this is one of the places where an honest article has to say so.

If you were hoping for a clean answer here, there is not one. What follows is the argument as it actually stands.

Eliseu Paglioli and colleagues reported better memory outcomes with the selective approach in hippocampal sclerosis [64]. Hans Clusmann and colleagues found outcome varied with resection type [65]. Johannes Schramm reviewed the whole question of optimal extent and concluded the field did not have a clean answer [66].

Then in 2013 Colin Josephson and colleagues ran a systematic review and meta analysis comparing standard against selective surgery and found the picture less favourable to the selective approach than its advocates hoped [67]. A 2026 study following patients for ten years compared temporal lobectomy against selective surgery on cognitive outcome [68], and another found the extent of hippocampal resection relates to neuropsychological outcome in a graded way rather than an all or nothing one [69].

The honest summary is a trade off, not a solution. Take less tissue and you may preserve more memory and control fewer seizures. Take more and the reverse. Where the balance sits depends on the individual brain in front of the surgeon, which is precisely why Chelune's question came to dominate.

Laser ablation is the newest attempt at the same idea. A thin fibre is guided into the target and heats it under MRI control, with no craniotomy. Daniel Drane and colleagues reported better object recognition and naming outcomes than open resection [70]. Joon Kang and colleagues and Robert Gross and colleagues reported on feasibility and outcomes [71][72]. A multicentre study of 234 patients examined how surgical targeting affects results [73].

Seizure freedom rates with ablation are generally somewhat lower than with open resection. The cognitive profile is generally better. That is the trade again, in newer equipment.

It is also worth reporting a clean negative result, because negative results are where a field is being honest. In 2026 Felix Zahnert and colleagues tested whether resecting a specific subregion of the piriform cortex predicted seizure freedom across two cohorts and found no link [74]. Not everything that looks like it should matter does.

Two further things the temporal lobe carries that rarely make it into consent conversations. LaBar and colleagues showed that fear conditioning is impaired after unilateral temporal lobectomy [75]. And for patients who are not surgical candidates at all, responsive neurostimulation offers a different route entirely [76].

Notice how different that is from the way surgery gets discussed publicly. The interesting question is almost never whether an operation works. It is what it costs, who pays that cost, and whether anybody measured it afterwards.

Modern series continue to track how these operations perform on seizures, cognition and quality of life together [77][78][79][80], including differences between men and women in pre and postoperative function that had gone largely unexamined [81].

The population itself is shifting. Classic hippocampal sclerosis is becoming less dominant in surgical series than it was [82], and late onset temporal lobe epilepsy has its own cognitive and biomarker signature [83]. That last point leads directly to the most under diagnosed thing in this entire article.

The Condition Almost Nobody Diagnoses

Imagine you are 62. You start having short episodes, usually on waking, in which you cannot remember what just happened. They last half an hour. Your family says you asked the same question four times. Then it passes completely and you are entirely yourself.

You also notice something else, and it is harder to describe. You remember a conversation perfectly on the day. A week later it is not there at all. Not vague. Gone.

You go to a doctor. You get a memory test. You pass.

That is transient epileptic amnesia, and it is a seizure disorder.

Nothing in that description sounds like epilepsy to most people, which is exactly why it goes unrecognised. There is no convulsion, no loss of consciousness, no dramatic event. There is a person who cannot hold on to a week.

Hodges and Warlow first sorted the transient amnesias into categories in a study of 153 cases in 1990 [84]. Kapur reformulated the epileptic variety in 1993 [85], and Zeman and colleagues described ten cases in detail in 1998 [86].

The paper that established it as a syndrome came in 2007. Christopher Butler and colleagues recruited 50 patients over eighteen months and compared them against 24 matched controls [87]. The numbers from that study are worth reading slowly.

Mean age of onset: 62 years. Episodes: a median of 12 per year, lasting a median of 30 to 60 minutes, occurring on waking in 37 of the 50 cases. Epilepsy was the initial specialist diagnosis in only 12 of the 50. And attacks ceased on antiseizure medication in 44 of the 47 patients who were treated.

Twelve out of fifty diagnosed correctly first time. Forty four out of forty seven treated successfully.

Put those two figures next to each other. The treatment is not the hard part. The recognising is.

A larger combined series followed in 2021, reporting 65 consecutive new cases alongside the earlier cohort of 50 and 102 cases from the literature [88]. Amnesia was the only manifestation of epilepsy in 24 percent of patients. Olfactory hallucinations occurred in 43 percent, motor automatisms in 41 percent, brief unresponsiveness in 39 percent. Epileptiform changes appeared on EEG in only 35 percent, and a suspected causative abnormality on MRI in only 5 percent. Seizures stopped with treatment in 93 percent.

Look at the last two bars. A normal MRI and a normal EEG are the rule, not the exception. If a clinician is waiting for a scan to show something before considering epilepsy, they will wait forever in 95 percent of these patients.

The disorder is still being missed in 2026, which is why a paper published that year is titled simply as a forgotten diagnosis [89]. Case reports continue to describe patients who progressed while partially treated [90].

You Can Pass the Test and Still Have Lost It

The three interictal features of transient epileptic amnesia are the reason it deserves a section of its own, and the first of them is the most useful thing in this article.

It is called accelerated long term forgetting. You learn material normally. You are tested at thirty minutes and you score normally. You are tested again a week later and it has vanished at a rate that no control participant shows.

Standard clinical memory tests are administered at about thirty minutes. That is the entire problem. The test is looking at the wrong point on the curve.

Think about how that feels from the inside. You know something is wrong. You are tested. You are told you are fine. The gap between your experience and the result is itself distressing, and it is not in your head.

Butler and Zeman laid out the pattern in 2008 alongside the other two features: autobiographical amnesia, in which whole stretches of a person's own past become inaccessible, and topographical amnesia, in which familiar routes and places stop feeling familiar [91]. Regional brain atrophy in these patients relates to the memory deficits [92]. Manes and colleagues documented the autobiographical component in detail [93][94], and Milton and colleagues characterised the remote memory deficits [95]. Metabolic imaging adds another layer [96]. Thorsten Bartsch and Christopher Butler wrote the review that places the whole family of transient amnesic syndromes side by side [97].

Is accelerated forgetting a failure to acquire the memory properly in the first place, or a failure to consolidate it afterwards? Serge Hoefeijzers and colleagues put that question directly in 2013 and did not resolve it [98]. It remains open.

Either way the practical consequence is identical. If you are tested at the only interval where your performance is normal, the result on the page says you are fine.

What is not open is how far the phenomenon extends. It occurs in temporal lobe epilepsy more broadly, across verbal, non verbal and autobiographical material [99]. It occurs in people with newly diagnosed focal epilepsy, before years of seizures have accumulated [100]. It occurs in children with generalised epilepsy [101]. Reviews now treat it as a recognised clinical entity in its own right rather than a curiosity [102][103][104].

And then it gets larger still. Alfie Wearn and colleagues found accelerated long term forgetting in healthy older adults, where it predicted cognitive decline over the following year [105]. In 2025 it was reported as a predictor of clinical onset in people carrying autosomal dominant Alzheimer's mutations, before symptoms appeared [106].

A test administered at the wrong interval may be missing early disease in more than one condition.

That is a large claim and it is being actively worked on rather than established. Watch it.

Which brings the story all the way back around. In 2024 a group in Edinburgh asked the obvious question that had gone unasked for seventy years: did Henry Molaison himself show accelerated long term forgetting [107]? The man whose case defined amnesia, re examined with the measurement tool his own case eventually made possible.

Why Sleep Keeps Appearing in This Story

If forgetting happens over days rather than minutes, the obvious suspect is what the brain does at night.

A memory is not finished when you form it. It gets worked on afterwards, and most of that work happens while you are asleep.

During slow wave sleep the hippocampus produces very fast bursts of coordinated activity called sharp wave ripples, which replay recent experience to the cortex at high speed. Gyorgy Buzsaki has argued they are the physical event underlying consolidation [108]. Bernhard Staresina and colleagues showed how they nest inside slower rhythms in the human hippocampus during sleep [109], and Axmacher and colleagues linked ripples in the human medial temporal lobe directly to memory consolidation [110].

That is the machinery. Your brain does not finish learning something when you stop studying it. It finishes hours later, in the dark, without asking you.

Now recall what interictal discharges do. They hijack the hippocampal to cortical coupling channel [37]. The same channel.

That is a mechanistically satisfying story and it should be labelled as a hypothesis rather than a demonstrated chain. The systems consolidation account it rests on has itself been challenged. Andrew Yonelinas and colleagues proposed a contextual binding theory in 2019 that reconsiders the standard model substantially [111]. Whether accelerated forgetting is a sleep problem, a daytime consolidation problem or an encoding problem is exactly the question Hoefeijzers left open.

There is a wider point here about how the hippocampus keeps similar memories from collapsing into each other. A structure that is being repeatedly interrupted cannot do that job cleanly, and the result is not dramatic amnesia. It is a slow smearing of detail that looks, from the outside, like ordinary forgetfulness.

What They Are Looking at Now

Research on memory in epilepsy in 2025 and 2026 is not chasing the hippocampus alone any more. Several directions are worth knowing about, and all of them should be read as current lines of enquiry rather than settled conclusions.

The glymphatic system, the brain's fluid clearance network, appears to function differently in temporal lobe epilepsy subtypes and the differences correlate with memory performance [112][113]. Basal forebrain cholinergic changes have been implicated in cognitive impairment in this population [114]. Pathological hypersynchronisation between hippocampus and cortex has been proposed as a direct mechanism of memory disruption [115]. Working memory impairment in right sided epilepsy has been traced to a specific hippocampal subfield [116]. A systematic review has drawn the cognitive and metabolic findings together [117].

Testing is also getting more honest about what everyday memory actually consists of. Relational memory, the binding of items to each other rather than items alone, is differentially impaired [118]. Autobiographical remembering and future thinking are both affected, which is the Hassabis finding from 2007 reappearing in a clinical population two decades later [119]. Functional MRI paradigms for lateralising memory are being refined [120][121], and the relationship between specific memory tasks and specific regional volumes is being tested rather than assumed [122][123]. Comparisons against frontal lobe epilepsy help separate what is specific to the temporal lobe from what is general to epilepsy [124]. Cognitive phenotypes are finally being described outside high income settings [125], and the neuropsychological profile differs by the histopathological type of sclerosis a patient has [126].

You do not need to follow any of those threads individually. What matters is the direction. The question has moved from where is memory to what exactly is going wrong in this particular brain, and that is a much harder question that could not even be asked seventy years ago.

Compare that to 1953. A surgeon with an estimate and no imaging, operating on both sides at once.

What Henry Molaison Could Not Consent To

There is a version of this story that treats him as a specimen. Most retellings drift into it without noticing, because the findings are so remarkable that the person producing them fades out.

He was a man who lived until 2008. He did crossword puzzles, which he could sometimes complete using knowledge acquired before 1953 [19]. Suzanne Corkin, who worked with him for close to fifty years, described someone patient and good humoured with the researchers he was introduced to every single time.

Consider that last detail for a moment. He was studied for fifty five years by people he met, from his point of view, for the first time on every occasion. He could not remember agreeing to any of it. He could not withdraw, because withdrawing requires remembering that you are enrolled.

Consent in this case was managed by a court appointed conservator. The arrangement was legally sound and it has been debated ever since, because the ordinary architecture of consent assumes a person who can carry a decision forward in time. He could not carry anything forward in time. That was the entire point of studying him.

Anaesthesia offers a mild and reversible version of the same discomfort: a person who is present, responsive and later has no record of it. Molaison lived in the permanent form.

He was also protected in ways that mattered. His identity was kept private for his entire life. He was known to the world as H.M. and became Henry Molaison only after he died, which is why you will see both names used and why this article mostly uses the one he was born with.

The scientific debt is enormous. So is the other kind.

What This Means If Memory and Epilepsy Overlap in Your Life

No advice here, because advice about your own brain has to come from someone who has examined it. What follows is only what the evidence in this article establishes.

Memory difficulty in epilepsy is common, it has several causes at once, and it is not the same thing as dementia. Seizure control is associated with recovery of function, not merely with prevention of further loss [28]. Surgery for drug resistant temporal lobe epilepsy has randomised evidence behind it and a real memory cost that varies enormously by which side and how much [46]. The risk is highest for people whose memory is best beforehand [51].

And a normal memory test at thirty minutes does not rule out a memory disorder, because at least one real and treatable condition is specifically invisible at that interval [87].

Those are facts about a literature. What they mean for one person is a conversation with a neurologist and a neuropsychologist, and nothing written on the internet substitutes for it.

Conclusion

In 1953 the field believed memory was everywhere in the brain. One operation on one man with epilepsy proved it was somewhere in particular, and the price of the proof was his entire remaining life.

In 2014 someone finally looked at the tissue and found the field had been teaching the wrong anatomy for half a century. The hippocampus was partly still there. The gateway was gone.

In between, epilepsy surgery learned to ask a different question. Not what is diseased, but what can be spared, and what will the rest of the brain manage without. That question has kept a great many people from repeating what happened in Hartford.

And in the clinic there is still a memory disorder that a standard test cannot see, in people around 62 who are told they are just getting older, that stops in more than nine cases out of ten when it is treated.

Henry Molaison is the reason we know that memory is a job done by a place. The unfinished part of his story is that we are still working out which place, and what it is really doing.

He never knew any of it. He was told many times.

Every time, for the first time.