How Memory Works: The Neuroscience of Why You Remember Some Things and Forget Everything Else
Henry Molaison had his hippocampus removed and could no longer form new memories. What that tragedy taught us about how memory actually works — and how unreliable it is.
Admin
Author
The man who could not remember tomorrow
In 1953, a twenty-seven-year-old man named Henry Molaison underwent experimental brain surgery to treat his debilitating epilepsy. The surgeon, William Beecher Scoville, removed the hippocampus and surrounding medial temporal lobe tissue from both hemispheres of Molaison's brain. The epilepsy improved. But something catastrophic had happened.
Henry Molaison — known to science for decades as "H.M." to protect his privacy — woke from surgery unable to form any new long-term memories. He could hold a thought in mind for a few minutes, but nothing transferred to permanent storage. He could remember his childhood, his early life, his personality. But every new person he met, every conversation he had, every meal he ate — all of it vanished. He lived in a permanent present tense until his death in 2008, unable to recognise his doctors even after meeting them hundreds of times. He was, as his neurologist Suzanne Corkin described, the most studied individual in the history of neuroscience.
H.M.'s tragedy gave us the single most important insight in the study of memory: memory is not one thing. It is a collection of distinct systems, involving different brain structures, that can be damaged independently of each other.
The architecture of memory
Sensory memory is the briefest: raw sensory input held for a fraction of a second to a few seconds while your brain decides what is worth processing further. Most of it is discarded immediately.
Working memory — sometimes called short-term memory — is the conscious workspace where you hold and manipulate information in the moment. It is what you use when you dial a phone number you just looked up, or hold the first half of this sentence in mind while reading the second half. Its capacity is famously limited.
In 1956, psychologist George Miller published a paper called "The Magical Number Seven, Plus or Minus Two," arguing that the human working memory can hold approximately seven chunks of information simultaneously — give or take two, depending on the individual and the type of material. This figure has been somewhat revised downward by subsequent research (modern estimates often suggest four chunks is more accurate), but the principle remains: working memory is severely limited, and this limitation is fundamental to how we think.
Long-term memory is the vast, largely unconscious system that stores information across days, years, and decades. It is itself divided into multiple sub-systems. Declarative or explicit memory stores facts (semantic memory: Paris is the capital of France) and personal experiences (episodic memory: your last birthday). Non-declarative or implicit memory stores skills, habits, and conditioned responses — things you know how to do without consciously knowing that you know them. This distinction explains one of H.M.'s most striking findings: despite being unable to form new explicit memories, he could learn new motor skills. If taught a mirror-drawing task each day, he improved every session — yet each day he would report having never done the task before.
From encoding to consolidation to retrieval
Memory is not a single event but a three-stage process.
Encoding is the initial registration of information. Not everything that enters your senses is encoded; attention is the gateway. If you are not paying attention, encoding is weak or absent. This is why you forget names immediately after being introduced to someone — social anxiety and divided attention prevent proper encoding from occurring in the first place.
Consolidation is the stabilisation of a memory over time, transforming it from a fragile, easily disrupted trace into a durable long-term store. Consolidation happens in two phases: early cellular consolidation (over hours) and systems-level consolidation (over days to years), in which memories are gradually transferred from hippocampus-dependent storage to more distributed cortical networks.
Sleep is critical to consolidation — a point discussed elsewhere, but worth noting here that the well-documented "sleep on it" wisdom is not metaphor. Slow-wave sleep replays hippocampal activity patterns from the day, apparently re-transferring important information to the cortex. People tested on newly learned material perform significantly better after sleep than after the same period of wakefulness.
Retrieval is what most people mean when they talk about memory: the active reconstruction of a stored experience. The word "reconstruction" is crucial, and it lies at the heart of why memory is far less reliable than we assume.
Emotion and the amygdala: why you remember where you were
You almost certainly remember where you were when you heard about a major world event — a terrorist attack, a sudden death of someone famous, a personal crisis. You may remember your exact location, what you were doing, even what you were wearing. These are called flashbulb memories, and they feel extraordinarily vivid and reliable.
The reason lies in the amygdala, an almond-shaped structure sitting adjacent to the hippocampus that processes emotional responses, particularly fear and threat. When an event carries strong emotional significance, the amygdala activates and signals to the hippocampus: this matters, encode this strongly. Stress hormones including adrenaline and cortisol amplify this signal further.
The result is that emotionally charged events are encoded more deeply and retrieved more readily than neutral ones. This is adaptive — remembering the dangerous situation matters more than remembering what you ate for lunch on an ordinary Tuesday.
But there is a catch: vividness is not accuracy. Multiple studies have followed people who gave detailed accounts of their memories of major events immediately afterwards, then re-interviewed them months or years later. The memories changed substantially — yet subjects were entirely confident in the accuracy of their later, altered accounts. The amygdala makes memories feel certain. It does not make them be correct.
The unreliable witness: Elizabeth Loftus and false memory
No researcher has done more to demonstrate the malleability of memory than psychologist Elizabeth Loftus. In a series of elegant and disturbing experiments beginning in the 1970s, she showed that memories can be created, altered, and even wholly fabricated — and that the person experiencing the false memory has no way to distinguish it from a real one.
In one landmark study, participants watched film footage of a car accident. Afterwards, some were asked "How fast were the cars going when they smashed into each other?" while others were asked the same question with the word hit instead of smashed. Those who heard "smashed" estimated higher speeds and were more likely, a week later, to (incorrectly) report having seen broken glass at the scene.
A single word in a question changed what people remembered seeing.
Loftus later went further, implanting entirely false memories. Using a technique involving a fake family photograph and leading suggestions from a relative, she and colleagues persuaded a significant percentage of participants that they had experienced being lost in a shopping mall as a child — an event that had never happened. Participants not only "remembered" the event but elaborated it with vivid detail.
These findings have had profound legal consequences. Eyewitness testimony, once regarded as the gold standard of evidence, is now known to be among the most unreliable forms of evidence. The Innocence Project, which uses DNA evidence to exonerate wrongfully convicted people in the United States, has found that eyewitness misidentification played a role in approximately 70% of wrongful convictions subsequently overturned.
The Ebbinghaus curve: why cramming does not work
Hermann Ebbinghaus, a German psychologist working in the 1880s, conducted a heroic and tedious set of experiments on himself: he memorised lists of nonsense syllables, tested his recall at various intervals, then re-memorised them and measured how much effort re-learning required. His results produced the forgetting curve — a mathematical description of how rapidly memories fade without reinforcement.
The curve is steep. Without review, roughly 50% of newly learned information is lost within an hour, 70% within a day, and around 90% within a week. This is why cramming for an exam produces short-term performance and almost no long-term retention.
The solution Ebbinghaus himself identified is spaced repetition: reviewing material at increasing intervals — a day later, three days later, a week later, a month later — such that each review occurs just as forgetting begins. This technique exploits a phenomenon called the testing effect or retrieval practice effect: the act of actively retrieving a memory strengthens it more than passively re-reading the material. Flashcard systems like Anki are built on this principle and have strong research support.
Why you cannot forget the smell of your grandmother's kitchen
Smell occupies a uniquely privileged position in memory. The connection between specific smells and vivid, emotionally intense autobiographical memories — called the Proust phenomenon, after Marcel Proust's famous madeleine scene — is real and neurologically explicable.
Most sensory information travels through the thalamus before reaching higher brain areas. But olfactory signals take a different route: they project directly from the olfactory bulb to the amygdala and hippocampus without thalamic relay. This means smell bypasses the sensory gating that applies to vision and hearing and arrives directly at the memory and emotion centres. The result is a uniquely intimate connection between smells and emotional memories, one that remains robust even in patients with early Alzheimer's disease who have lost other forms of memory.
Techniques that actually improve memory
The evidence-based interventions are fewer and less glamorous than the brain-training industry would like. No app or supplement has demonstrated reliable, transferable memory improvement in healthy adults. But several approaches do work.
Spaced repetition beats massed practice every time for long-term retention. Elaborative encoding — actively connecting new information to existing knowledge rather than passive repetition — dramatically improves storage. Sleep before and after learning matters. Exercise increases hippocampal neurogenesis and is one of the few interventions shown to slow age-related memory decline. Testing yourself is more effective than re-reading.
And perhaps most importantly: reducing interference helps. The degree to which multitasking impairs encoding is substantial. Paying full attention during learning — unseductive and obvious advice in an age of notifications — remains one of the most evidence-supported things you can do for your memory.
The bottom line
Memory is not a recording device. It is a reconstruction system — creative, associative, and remarkably fallible. The hippocampus encodes, sleep consolidates, and the amygdala tags what matters. But retrieval rewrites the original, emotion inflates confidence without improving accuracy, and leading questions can implant experiences that never occurred. Understanding this does not diminish memory's wonder; it deepens it. And it offers a practical corrective to the confidence we place in our own recollections — and those of others.