In the 1880s a German psychologist named Hermann Ebbinghaus set out to measure something nobody had measured before: forgetting. To do it he needed things to learn that meant nothing to him, so that nothing he already knew could help, and so he made up syllables that weren’t words and wrote them in rows of thirteen. He read a row over and over until he could say it through without a mistake. Then he waited, twenty minutes, or a day, or a month, and learned the same row again.
He couldn’t simply count what was left, because most of the time he couldn’t say a single row back. But the second learning went quicker than the first, and the time it saved was a measure of what had survived. He was the only person in the experiment, and he kept at it for months.
In 2011 a student in Amsterdam did the whole thing again, and the results were published in 2015: 70 lists of syllables, about 70 hours of learning and relearning, over 75 days. His curve came out much like the one Ebbinghaus had drawn more than a century earlier.
The shape is the thing to see. Most of the loss comes almost at once, within the first hour, and then it slows to a crawl. A month later Ebbinghaus still saved about a fifth of the time it had taken him to learn the row. The repeat lost more by the end, partly because, as the weeks went on, its learner took longer to learn every new list. One small bump stands out at a day. The repeat did a little better after a full day than after nine hours, and Ebbinghaus’s own one-day score was higher than his curve expected. He didn’t trust it, tested it again, and got much the same. Some researchers think a night’s sleep is the reason; the team that repeated the study says that would need more research.
So forgetting is fast, then slow. But a curve doesn’t say why. The plain answer, which the rest of this chapter fills in, is that very little is ever simply wiped out. A memory can fail because it never went in properly, because it faded, because other memories got in its way, or because it is in there and can’t be found. And what you do find, you put together again each time, which is why a memory can feel certain and still be wrong.
Why things are lost
Forgetting is losing the ability to bring back something you once learned, for a while or for good. It is normal and it happens all the time. It is fastest just after learning. It is often partial: you know the name starts with s, or you can see the actor’s face and hear his voice. And it is often temporary, as when the name you were reaching for all afternoon comes to you at night. That feeling has a name of its own, being on the tip of the tongue.
Some of what feels like forgetting was never learned. In 1979 two American psychologists showed people drawings of a one-cent coin, the commonest coin in the country, and most couldn’t pick out the real one. They had handled thousands, but they had only ever stored enough to tell it from other coins. Psychologists call this encoding failure. Try drawing a coin or a note you use every week, then look at it.
For what was learned, there are two classic causes, and they are the two CDC’s course names. The first is trace decay. Learning, the idea goes, leaves a trace in the brain. What the trace is made of is thought to be stronger connections: when nerve cells signal to each other again and again, the connection between them becomes stronger and stays so, which is called long-term potentiation. A trace that isn’t used, the idea goes, fades with time, like a path across a field that nobody walks. The short-term memory experiment of the last chapter looks like that: three letters, then a few seconds of counting backward, and they are mostly gone.
The trouble is that time never passes empty. Those people were counting, and the counting filled their minds. In a classic test of this, people heard long lists of digits read fast or slowly and were asked about one of them. How many digits came after it mattered far more than how many seconds had passed. When experiments in 2009 were designed to stop people rehearsing, so that time alone could be measured, time on its own did very little to short-term memory, and other things to remember did much more. Decay seems to happen in sensory memory, the store that holds a glimpse for a second. Whether it also wipes long-term memories is hard to test, and it fits badly with the old song from childhood that comes back whole after twenty years of never being thought of.
The second cause is interference: memories get in each other’s way, and the more alike they are, the worse. It works in both directions. When old learning gets in the way of new, it is proactive interference. In Baishakh, when the new year starts, someone always writes last year’s date at the top of the page. When new learning gets in the way of old, it is retroactive interference: once you have used your new phone number for a year, the old one is hard to bring back. In experiments people learn pairs of words, cat with tree, and then a second list where cat goes with glass. Asked for the first list, they do worse than people who never learned the second one.
In 1957 an American psychologist looked back over the forgetting experiments and argued that most of their forgetting came from lists the people had learned before. Ebbinghaus had learned a huge number of rows of syllables, each one very like the others. Even the Amsterdam team wondered whether interference, or plain tiredness, was why their learner grew slower over the weeks. And in the 1920s, people who slept after learning remembered more than people who spent the same hours awake, which was read at the time as fewer new things getting in the way. Sleep, it turns out, does more for memory than that, which is part of the next chapter.
Last, a memory can be there and not come out, for want of the right cue. People who can’t recall a word from a list they learned will often bring it back when given a cue linked to it, such as the group it belonged to. This is the tip of the tongue again: the memory is available, but it can’t be reached from where you are.
All of that is everyday forgetting. Amnesia is something else, a loss of long-term memory caused by disease, an injury or a terrible experience. It comes in two kinds, named for the direction of the loss from the day of the harm. In anterograde amnesia a person can’t store new memories after it, often from a blow to the head. Henry Molaison, known as H.M., was left this way by brain surgery: he would read the same magazine again and again, each time as if for the first time, while still getting faster at a puzzle he practiced day after day. In retrograde amnesia a person loses memories from before the harm, as someone who wakes after a car accident may.
Rebuilt each time
In 1932 the British psychologist Frederic Bartlett, at Cambridge, published a study that changed how memory is thought about. He gave his English students a Native American folk tale called “The War of the Ghosts” and asked them to write it out from memory, again and again, over months and sometimes years.
The story drifted, and always in the same direction, toward the kind of story an English reader expects. Canoes became boats. Hunting seals became fishing. Strange names changed and then dropped out. Near the end of the tale a man dies at sunrise, and “something black came out of his mouth.” One reader’s first version kept it as a black substance. Nearly four months later, the same reader remembered the people seeing a ghost come out of his mouth, and then corrected it: really, it was “a kind of materialization of his breath.” That reader knew the phrase wasn’t in the story, and wrote it anyway.
Bartlett’s conclusion was that remembering isn’t like taking a copy off a shelf. You keep a few pieces and a general sense of the thing, and each time you remember, you build it again from those, filling the gaps with what you know about how such things usually go. He called that background knowledge a schema: your schema for a wedding, a bus ride, a ghost story. The built version is what you’ll remember next time. His methods were loose by today’s standards, but the idea has held: memory rebuilds, it doesn’t replay.
You can catch your own memory filling a gap. Try this before reading on.
Every word on that list was something you’d think of near the word doctor, and doctor itself was never shown. In 1995 two American psychologists brought back an old test like this, with lists of words linked to one missing word, such as bed, rest, awake, tired for sleep. Asked straight after a list to write down its words, people wrote the missing one 40% of the time in one experiment and 55% in another, and later picked it out as seen, with high confidence, about as often as words they had really seen. As the words go by, each one seems to call up its close neighbors in your mind, and at the test, the one most called up feels as if it was there. It is a small, harmless trick, but it is the same machinery as the big mistakes. These were American word lists tested on American students, so it may work less well on a reader in Nepal, and one reader’s answer proves nothing either way.
What other people say can get into a memory too. In 1974 the American psychologist Elizabeth Loftus and a colleague showed 45 students films of car crashes and asked how fast the cars were going when they “smashed into” each other, or “collided,” “bumped,” “hit” or “contacted.” The answers went down with the word, from about 41 miles an hour for smashed to about 32 for contacted. A week later the students were asked whether they had seen broken glass. There was none in the film, but the smashed group was more than twice as likely to say there was. The word in the question had become part of what they saw. This is called the misinformation effect.
A memory can also be true and pinned to the wrong source. The Australian psychologist Donald Thomson was once picked out by a woman as the man who had attacked her. He had a perfect alibi: at the time, he was on live television, talking about memory. She had been watching that program when she was attacked, and her memory joined the attacker to the face on the screen.
Could a whole event be planted? In 1995 Loftus and Jacqueline Pickrell asked people about four events from their childhood that, they were told, an older relative had described. Three really had happened. The fourth was made up with the relative’s help: being lost in a shopping mall at about the age of five. After two interviews, a quarter of the 24 people remembered at least part of being lost.
That study was small, and expert witnesses have cited it in court cases ever since, so in 2023 a team in Ireland repeated it with 123 adults, its methods and its counting set out in advance. Judged by the researchers from what people said, 35% had some memory of being lost, most of them only in parts. Asked at the end, 14% said themselves that they remembered it, and another 52% believed it had happened though they didn’t remember it. The made-up event was described in far fewer words, and less clearly, than the true ones. How to count all this is argued. In 2024 two other researchers went through the interviews and found that those judged to have a false memory recalled, on average, fewer than two of the six details they had been given, and that about half the people described experiences of their own that may really have happened. Counted that way, according to the replies, only 4% formed a false memory of the event. Loftus and others answered in 2025 that all memory is rebuilt from many sources and that even low rates matter; the Irish team disagreed with the 4%; and the two critics replied that their concerns remained.
Even a memory that feels burned in can drift. On 12 September 2001, the day after the attacks on New York, 54 students at an American university wrote down how they had heard the news, and an ordinary event of the same days. Weeks or months later they wrote it all again. Both memories changed as much as each other. But the students stayed far surer of the attack memory, and saw it more vividly. A strong feeling makes a memory feel right; it doesn’t make it more accurate.
Repetition works on memory in another way: a claim you have met before feels more familiar, and familiar feels true. That, and how easily recalled examples sway judgments, are told in Thinking Clearly, in The feed and Shortcuts.
When it matters
In 1984 a 22-year-old college student in North Carolina, in the United States, was raped. Her name is Jennifer Thompson (now Thompson-Cannino), and she has told the story publicly ever since. Shown six photographs by the police, she looked for four or five minutes before she said of one, “Yeah. This is the one,” and then, “I think this is the guy.” She asked whether she had done all right, and the detective told her she had done great. Later, unsure between two men in a line of seven, she picked the same man, Ronald Cotton, because he looked most like him. By the time of the trial she had no doubt at all, and he was sent to prison for life. In 1995 a DNA test showed that the attacker was another man, already in prison, who had admitted the crime to other prisoners. Cotton had served more than ten years. After he was freed, the two met, became friends and wrote a book together.
Thompson did nothing wrong. She tried hard to remember, and her memory failed in the ordinary ways this chapter has described. When the Innocence Project, an American group, looked at the first 239 people freed by DNA evidence, a mistaken eyewitness had played a part in about three convictions in four.
This doesn’t mean a witness should never be believed, and that matters just as much. A careful review in 2017 found that when a witness is tested once, early, in a fair line-up run by someone who doesn’t know which person is the suspect, is told the culprit may not be in it, and says straight away how sure they are, a confident choice is usually right, and a hesitant one is often wrong. When the conditions aren’t like that, confidence can mislead, and confidence that grows later, after a hesitant first pick and a word of praise, as Thompson’s did, is no guide at all. So some police forces now have line-ups run by an officer who doesn’t know which one is the suspect, and in some American states judges warn juries about mistaken eyewitnesses.
The hardest argument about memory is about what a mind does with pain. Sigmund Freud thought the mind protects itself by pushing unbearable memories out of awareness without knowing it is doing so, and he called this repression. Pushing a memory away on purpose, by not letting yourself think of it, is called suppression. Repression is one of the most famous of Freud’s ideas, but most of his ideas haven’t found support in modern research, and this one is still argued over.
In the 1990s the argument became bitter. Some therapists used hypnosis, guided imagining and age regression to help patients find memories of abuse they believed were hidden. Loftus and others pointed to studies like the ones above and asked whether some of those memories had been built in therapy.
Out of that came a statement by the American Psychological Association on what memory researchers and clinicians agree on. Most people who were sexually abused as children remember all or part of what happened, though they may not fully understand it or tell anyone. Rarely, a memory of abuse that was forgotten can come back later. It is also possible to build a convincing memory of something that never happened. And without other evidence, nobody can tell the two apart, the person remembering included.
Both halves have evidence behind them. In one study, 129 women whose abuse as children had been reported 17 years earlier were interviewed in detail, and 38% didn’t recall the abuse that had been reported. Those who were younger at the time, or abused by someone they knew, were more likely to have no memory of it. Long gaps in memory, the researcher concluded, are no evidence that abuse didn’t happen. In another study, the memories of people who had suddenly remembered abuse outside therapy could be backed up by other evidence about as often as those of people who had never forgotten, while memories recovered in therapy were backed up less often. Some clinicians think the mind can lock such a memory away for years; many researchers say there is little evidence for that, whatever the reason the memory went.
None of this is a reason to doubt someone who tells you they were harmed. Most people remember what was done to them, many take years to say it, and remembering late is no sign of making it up. If a memory troubles you, tell someone you trust.
What to carry away is simpler than all the argument. Forgetting is mostly not erasing: things never go in, get in each other’s way, or can’t be found, and the loss is fast at first and slow after. And remembering is rebuilding: each time, you put the past together from pieces and from what you know, and words, suggestions and later knowledge can slip into the building without your feeling a thing. That is why a memory can be vivid, confident and wrong, and why the people who study memory most are the slowest to say “I remember it exactly.”
Found a mistake? Please tell us.
