Your memory rebuilds the past every time you recall it

Ask two siblings about the same childhood holiday and you can get two confident, detailed and incompatible stories. Usually nothing rides on it. Sometimes a great deal does.

The Innocence Project, a US legal organization, tallied the first 375 people in the United States cleared by DNA evidence between 1989 and 2020. Eyewitness misidentification played a part in 69% of those wrongful convictions. Most of those witnesses were not lying. They remembered, sincerely and wrongly.

That is the central fact about human memory, and it surprises people. Your brain does not save the past like a video file and press play. It rebuilds each memory when you call it up, from fragments, expectations and whatever you have heard since. That design works well most of the time. It also fails in ways researchers can measure.

Memory is several systems, not one storehouse

Much of what scientists know about memory’s architecture began with one man, known in the research literature as H.M. In 1953, at age 27, he had experimental surgery on both sides of his brain to control severe seizures. The seizures eased. His ability to form new memories of facts and events largely vanished.

Researchers studied him for five decades, until his death in December 2008, and a 2009 review in the journal Neuron by Larry Squire of the Veterans Affairs system in San Diego lays out what they learned.

H.M. could still repeat back six or seven digits, a normal span, and could hold a three-digit number for up to 15 minutes by rehearsing it constantly. The moment his attention shifted, it was gone. Faces and abstract designs, which are hard to repeat to yourself, slipped away in under a minute.

The real surprise came from a skill test. Psychologist Brenda Milner had him trace a star while seeing his hand only in a mirror. Over ten attempts he got better, and he kept the improvement across three days. Yet at the end of testing, he had no recollection of ever having done it.

That split became a founding idea of the field. Declarative memory is what most people mean by memory: facts and events you can consciously call up. Nondeclarative memory is an umbrella for skills, habits, simple conditioning, emotional learning and priming, and it leans on other structures, including the basal ganglia, the cerebellum and the amygdala. A 2004 review by Squire describes the modern picture as multiple separate memory systems rather than a single faculty.

Two popular retellings of H.M.’s story need correcting:

• “He proved the hippocampus is the memory center.” His surgery removed the hippocampus plus the amygdala and neighboring cortex, so his case alone could never single out the hippocampus. MRI scans later showed the damage reached a little more than 5 centimeters back, not the 8 the surgeon had described.
• “He could never learn anything new.” He learned a motor skill, and Squire notes that he picked up significant new knowledge over the years. He also recognized faces of people who were famous before his operation as well as, or better than, people his own age, a sign that the damaged region is not where long-held memories are finally kept.

How a memory gets made, kept and found

Researchers break remembering into three stages. Encoding is getting information in. Storage is keeping it. Retrieval is getting it back out when you need it. Each can fail separately: “I never noticed” and “it’s on the tip of my tongue” are different problems.

The National Institute of Neurological Disorders and Stroke describes the hippocampus, a small structure in the brain’s medial temporal lobe, as a kind of indexer. It sends memories out to the right parts of the brain’s two large hemispheres for long-term storage and helps pull them back when needed. The same agency describes the frontal lobes as short-term holding areas that let you keep one idea in mind while weighing others, and links the cerebellum to learned movements such as hitting a tennis ball.

The holding area is smaller than you’ve heard

Many people have heard that the mind can juggle “seven, plus or minus two” items, a figure drawn from a famous 1956 paper by the psychologist Miller. In a 2001 review in Behavioral and Brain Sciences, N. Cowan of the University of Missouri argued that Miller’s seven was meant more as a ballpark figure and a turn of phrase than as a hard limit. Pulling together many kinds of experiments, Cowan put the central capacity at about four chunks, with pure estimates falling between three and five.

A chunk is anything you already treat as one unit. Grouping a phone number into three parts is chunking. Cowan’s limit shows up when people are prevented from grouping or rehearsing, which is also why those everyday tricks help so much.

What the famous forgetting curve really measured

The best-known graph in memory research comes from Ebbinghaus, who in the 1880s memorized lists of meaningless syllables, using only himself as the subject. He measured how much faster he could relearn each list after delays ranging from 20 minutes to a month. His steep early drop is often retold as “you forget most of what you learn within a day.”

That is not what he measured. His yardstick was savings: how much less time relearning took than first learning. In 2015 two University of Amsterdam researchers, Jaap Murre and Joeri Dros, repeated the experiment with Dros as the single subject, spending 70 hours learning and relearning lists of 13 nonsense syllables. The curve came out close to the original, so the classic finding does replicate.

The analysis also showed why the popular version misleads. Scored as the share of syllables recalled correctly, the same data produced curves much shallower than the savings curves. Position mattered, too: the last few syllables in each list were barely forgotten at all. One person learning nonsense is a thin basis for predicting how much of a lecture or a discharge instruction sheet you’ll keep. The authors also saw a probable upward jump at the one-day mark that some researchers attribute to sleep, a question that belongs to the science of sleep and learning rather than to this article.

Recall is reconstruction, and one word can change it

In a 1974 study still taught in introductory courses, Elizabeth Loftus and her colleague J. C. Palmer showed 45 students seven short films of traffic accidents and asked how fast the cars were going. Only one word in the question changed. Students asked how fast the cars were going when they “smashed” into each other gave an average estimate of 40.8 mph. With “hit,” the average was 34.0. With “contacted,” it was 31.8.

The guesses weren’t accurate to begin with. In four of the films the real speeds were known, and the collision that really happened at 20 mph drew an average estimate of 37.7 mph, nearly double.

A second experiment went further. A week after watching a crash film, 150 students were asked whether they had seen any broken glass. The film showed none. Among students first questioned with “smashed,” 16 of 50 said yes, compared with 7 of 50 questioned with “hit” and 6 of 50 in a comparison group. The verb had not just tilted a number. It had added a detail to the memory itself.

Loftus’s 2005 review in Learning & Memory calls this the misinformation effect: memory for an event gets worse after exposure to misleading information about it. The review describes three decades of studies showing the effect, notes that some people are more susceptible than others, and reports that warning people about misinformation helps only in limited circumstances.

A fair note on replication. In 2017 a team based at the University of San Diego tried the idea with a hockey collision. Written verbs made no difference to estimated skating speed, and when the verbs came through live commentary, “contact” produced higher estimates than “smash.” That was a conceptual replication in a new setting, not a repeat of the car study, so it limits how far the specific verb effect travels without overturning decades of misinformation research.

Can you remember something that never happened?

The best-known test of that question is the “lost in the mall” study. In 1995 Loftus and her colleague Pickrell gave volunteers booklets describing four events from their childhood. Three were real, supplied by an older relative. The fourth, getting lost in a shopping mall, was invented. Six of 24 participants, a quarter, came to report some memory of it.

Twenty-four people is a small sample, and the study has been cited in courtrooms. So in 2023 a team led by Gillian Murphy at University College Cork ran a larger, preregistered replication, meaning the methods and the definition of success were registered before any data came in. With 123 adults aged 18 to 57, the coded false-memory rate climbed through the study: 15% after reading the booklet, 19% at a first interview and 35% by the second interview, made up of 33 partial and 10 full false memories. Asked directly at the end, 14% said they remembered being lost in the mall. Another 52% said they didn’t remember it but believed it had happened.

Then came the argument. Bernice Andrews of Royal Holloway and Chris R. Brewin of University College London, both emeritus psychology professors, reanalyzed the publicly released interview transcripts in the journal Applied Cognitive Psychology. Writing in Scientific American in 2025, they argued that only five participants could reasonably be said to hold a false memory, not the 43 originally counted. People coded as having a full false memory recalled fewer than three of the story’s six core details on average. Half of those coded had been lost before or had a similar experience, which they may have been describing.

A team including Kimberley A. Wade and Loftus replied in Applied Cognitive Psychology in January 2025. Their position: memory reports naturally blend accurate fragments, general knowledge and guesswork, and suggestion can still produce substantial distortions, so even a low rate of planted memories matters. Brewin and Andrews answered in September 2025 that their concerns about the method’s reliability, and about false-memory findings being overstated in expert testimony, still stand.

Both camps agree that suggestion can distort memory. They disagree about how often a complete memory of an invented event can be planted. On this library’s evidence scale, the claim that suggestion can plant whole events rates as positive evidence, with the size of the effect still contested. Anyone who says “a quarter of people” can be given a false childhood memory is skipping the argument, and so is anyone who calls false memories a myth.

Even extraordinary memories bend

Some people do have exceptional memories. In 2012 researchers at the University of California, Irvine, described 11 people with highly superior autobiographical memory, or HSAM. They could recall public and personal events, and the dates those events happened, far better than comparison volunteers. On most standard laboratory memory tests, though, they scored about the same as people of the same age and sex.

A 2013 study in PNAS then asked whether such people are immune to false memories. They are not. Like the comparison group, they falsely recognized words that had never been on a studied list. They also showed more false memory than controls on one misinformation task, and were just as likely to say they had seen footage of a plane crash that does not exist. The authors, with Lawrence Patihis as first author, concluded: “Finding false memories in a superior-memory group suggests that malleable reconstructive mechanisms may be fundamental to episodic remembering.”

So the idea of a photographic memory that records and replays without distortion gets no support from this work. Even the most remarkable autobiographical memories on record are rebuilt, not replayed.

Confidence is not the same as accuracy

The Innocence Project’s review of those 375 DNA exonerations shows how ordinary memory errors travel through a justice system. Among the cases involving misidentification, 84% involved a surviving victim, 77% involved more than one identification procedure, and 42% crossed racial lines. People in the full set served 14 years on average, and 21 had spent time on death row.

In 2014 the National Research Council of the National Academies reviewed the science in a consensus report, Identifying the Culprit. One line in its findings sums up the problem: “Unknown to the individual, memories are forgotten, reconstructed, updated, and distorted.” The committee found that a witness’s confidence at trial is not a reliable predictor of accuracy, while confidence at the first identification may track accuracy better. Its 11 recommendations include double-blind lineups, in which the officer running the lineup doesn’t know who the suspect is; standard instructions for witnesses; and recording, in the witness’s own words, how confident they are at that first identification.

The US Department of Justice set similar rules for photo arrays in a January 2017 memo from Deputy Attorney General Sally Q. Yates. It calls for blind or “blinded” administration, at least five filler photos alongside the one suspect, telling witnesses the culprit may not be pictured, no feedback on their choice, and a record of the witness’s confidence statement. Its appendix explains why feedback matters: a confirming remark from an investigator can make a witness surer without making them more accurate. The memo took no position on showing photos one at a time versus all together, because the research had not settled that question.

Hypnosis turned up in 2% of the misidentification cases; why hypnotically refreshed recall is so risky is the subject of our hypnosis article. The broader lesson here: confidence captured early, before anyone has nudged the witness, carries real information. Confidence that has grown over months of retelling carries much less.

Protecting the memories that matter

You can’t switch reconstruction off, but you can work with it.

• Write it down early. After a car crash, a fall, a frightening symptom or a tense meeting, note what you saw and heard as soon as you safely can, before conversations and news coverage reshape it.
• Tag where each detail came from. Mark anything that came from another person, a photo or an adjuster’s question. Months later, borrowed details feel exactly like your own.
• Notice how questions are worded. “How fast was he going when he slammed into you?” already contains an answer. When you’re helping someone else remember, open questions such as “What happened next?” do less damage.
• Keep a health timeline. When symptoms began, which medicines changed and what a test showed are exactly the details memory rebuilds poorly. A dated list makes a clinic visit more accurate than recall alone.
• Treat family stories as stories. Two relatives can remember the same event differently and both be sincere. Old photos, letters and calendars settle more than arguing does.
• State your confidence once, at the start. If you ever identify someone, say how sure you are in your own words at that moment and ask that it be recorded.

When memory trouble is a medical question

Ordinary reconstruction errors are not an illness. A sudden change is different. The Centers for Disease Control and Prevention lists sudden confusion, trouble speaking or understanding speech, numbness or weakness on one side, trouble seeing or walking, and a severe headache with no known cause as stroke signs. Call 911 right away, note when symptoms began, and wait for an ambulance rather than driving to the hospital.

Memory loss after a head injury, or memory trouble that shows up around a new prescription or a change in drinking, also deserves a clinician’s attention. MedlinePlus, from the National Library of Medicine, lists concussion, stroke, alcohol, prescription and illegal drugs, low vitamin B1 or B12, depression and dementias among the possible causes, and notes that a clinician will ask whether the problem came on suddenly or slowly. If you suspect a medicine, the person who prescribed it is the one to ask whether the timing fits. How to tell everyday age-related slowing from a change that needs testing is covered in our pieces on memory change and assessment.

The most useful thing to remember about memory

A memory that feels vivid, detailed and certain can still be wrong, and you cannot tell from the inside. The work on H.M., on leading questions, on implanted events and on people with near-perfect autobiographical recall points the same way: remembering is reconstruction.

So when accuracy matters, at a clinic, after an accident or in a courtroom, give the most weight to records made early, keep track of where each detail came from, and trust what you wrote down before anyone else’s version reached you.

Keep exploring: smartphone brain drain and memory screening test.

Sources

1. Innocence Project. “DNA Exonerations in the United States (1989 – 2020).” innocenceproject.org
2. National Institute of Neurological Disorders and Stroke (NIH). “Brain Basics: Know Your Brain.” ninds.nih.gov
3. Neuron. “Squire LR. "The legacy of patient H.M. for neuroscience." Neuron, 2009. PubMed. DOI: 10.1016/j.neuron.2008.12.023.” pubmed.ncbi.nlm.nih.gov
4. Neurobiology of Learning and Memory. “Squire LR. "Memory systems of the brain: a brief history and current perspective." Neurobiol Learn Mem, 2004. PubMed. DOI: 10.1016/j.nlm.2004.06.005.” pubmed.ncbi.nlm.nih.gov
5. Behavioral and Brain Sciences. “Cowan N. "The magical number 4 in short-term memory: a reconsideration of mental storage capacity." Behav Brain Sci, 2001. PubMed. DOI: 10.1017/s0140525x01003922.” pubmed.ncbi.nlm.nih.gov
6. PLoS One. “Murre JM, Dros J. "Replication and Analysis of Ebbinghaus' Forgetting Curve." PLoS One, 2015. PubMed. DOI: 10.1371/journal.pone.0120644.” pubmed.ncbi.nlm.nih.gov
7. Journal of Verbal Learning and Verbal Behavior. “Loftus EF, Palmer JC. "Reconstruction of automobile destruction: An example of the interaction between language and memory." J Verbal Learning Verbal Behav, 1974;13:585-589. DOI: 10.1016/S0022-5371(74)80011-3.” doi.org
8. Learning & Memory. “Loftus EF. "Planting misinformation in the human mind: a 30-year investigation of the malleability of memory." Learn Mem, 2005. PubMed. DOI: 10.1101/lm.94705.” pubmed.ncbi.nlm.nih.gov
9. Quarterly Journal of Experimental Psychology. “Goldschmied N et al. "Appraising Loftus and Palmer (1974) post-event information versus concurrent commentary in the context of sport." Q J Exp Psychol, 2017. PubMed. DOI: 10.1080/17470218.2016.1237980.” pubmed.ncbi.nlm.nih.gov
10. University College Cork (CORA repository). “Murphy G et al. "Lost in the Mall Again: A Preregistered Replication and Extension of Loftus & Pickrell (1995)." Memory, 2023. DOI: 10.1080/09658211.2023.2198327.” cora.ucc.ie
11. Scientific American. “Andrews B, Brewin CR. "People Likely Aren't as Susceptible to False Memories as Researchers Thought." Scientific American, July 2025. DOI: 10.1038/scientificamerican072025-31HTZLNuqsARs9ouH2Teqa.” scientificamerican.com
12. University College London. “Implanting false memories much harder than claimed in court (news release, 7 January 2025).” ucl.ac.uk
13. University of Warwick (WRAP repository). “Wade KA, Riesthuis P, Bücken C, Otgaar H, Loftus EF. "Still lost in the mall: false memories happen and that's what matters." Applied Cognitive Psychology, 2025;39(1):e70028. DOI: 10.1002/acp.70028.” wrap.warwick.ac.uk
14. University College London (UCL Discovery repository). “Brewin CR, Andrews B. "Major Concerns About the False Memory Implantation Paradigm Persist." Applied Cognitive Psychology, 2025;39(5):e70108. DOI: 10.1002/acp.70108.” discovery.ucl.ac.uk
15. Neurobiology of Learning and Memory. “LePort AK et al. "Behavioral and neuroanatomical investigation of Highly Superior Autobiographical Memory (HSAM)." Neurobiol Learn Mem, 2012. PubMed. DOI: 10.1016/j.nlm.2012.05.002.” pubmed.ncbi.nlm.nih.gov
16. Proceedings of the National Academy of Sciences. “Patihis L et al. "False memories in highly superior autobiographical memory individuals." PNAS, 2013. PubMed. DOI: 10.1073/pnas.1314373110.” pubmed.ncbi.nlm.nih.gov
17. National Academies of Sciences, Engineering, and Medicine (National Research Council). “Identifying the Culprit: Assessing Eyewitness Identification (2014), Chapter 6: Findings and Recommendations.” nationalacademies.org
18. U.S. Department of Justice, Office of the Deputy Attorney General. “Eyewitness Identification: Procedures for Conducting Photo Arrays (memorandum, January 6, 2017).” justice.gov
19. MedlinePlus (U.S. National Library of Medicine). “Memory loss.” medlineplus.gov
20. Centers for Disease Control and Prevention. “Signs and Symptoms of Stroke.” cdc.gov