How memory works
Is photographic memory real?
Not in the sense people mean. No one has been shown, in a test that held up when others tried it, to look at a page once and read it back later like a photograph. One person passed a demanding test of exact recall in 1970 and was never tested again. The real effect closest to it is eidetic imagery: a small share of children report still seeing a picture for a while after it goes, and what they report often contains errors. The memory athletes and trained subjects who have been studied use methods they practised.
You remember the answer was at the top of the left-hand page, and you can nearly see it there. Try reading the line underneath. The claim is about that second part: reading from the page.
- Popular term
- photographic memory, not a scientific one
- Closest real effect
- eidetic imagery (eye-DET-ik), from the Greek for visible form
- Who has it
- about 2% to 10% of preadolescents in one summary; virtually no adults
- Lasts
- half a minute to several minutes, by Searleman’s account
- Accurate?
- often not: details get altered and sometimes invented
- Trained memory
- one subject went from 7 to 79 digits in over 230 hours of practice
An AI drafted this page from the sources below and another AI checked it. No person has tried the demo yet.
Try it
Can you lay one picture on another?
A perfect copy of the first picture would make every size equally easy: you’d hold it and see the empty square when the second arrives. It’s a strict test, because it asks for exactly what the popular claim says.
Each round has two pictures of dots on one grid, one after the other. Laid on top of each other, they fill every square but one. Find that square.
- Look at the first picture for up to 30 seconds. Press Done when you’re ready.
- The grid goes blank for five seconds, then the second picture comes.
- Tap the blank square you think the first picture didn’t fill either.
There are 5 rounds, from 4 by 4 to 12 by 12, which takes up to about four minutes if you use all the time. No pen and no screenshots, since those are what the claim says you wouldn’t need.
This is our version of the fusion method, explained below. The five-second blank is there because iconic memory, the visual trace everyone has, is gone within about a second, and the fragile store some researchers place after it lasts about four.
It tests the strongest form of the claim. The smallest grid proves little, since each dot you do remember rules out a square and seven is close to the span people manage for digits. It can’t detect eidetic imagery as researchers define it, which is a reported image and often an inexact one, so missing the big grids doesn’t rule that out. Your result says how you did on these grids today, with no bearing on your memory in general. Your answers stay on this page, and we only count that a demo was started. The picture’s time on screen is measured by the page, and a round where the tab was hidden runs again.
If you can’t see it: this demo has no version that works without sight, because the claim it tests is about images. The diagram in the next section is described in text for screen readers.
What is eidetic imagery?
It’s the effect that research on photographic memory actually found. The usual test, Searleman writes, puts an unfamiliar picture on an easel for 30 seconds, takes it away, and asks the person to keep looking at the easel and say what they can observe. Someone with eidetic imagery says they still see it, scans it with their eyes, and describes it in the present tense.
A 1964 paper by R. N. Haber and R. B. Haber launched the modern research, Crowder says. They tested 150 elementary-school children, each shown four pictures. More than half, 84, reported some kind of image, and 12 of those stood clearly apart. Crowder gives the size of the gap: positive colours in an average of 90% for the twelve against 34% for the other 72 with some imagery, and scanning the blank easel in 100% against 2%.
| Survey | Who | Eidetic |
|---|---|---|
| Haber & Haber 1964 | 150 elementary-school children | 12 (8%) |
| Paivio & Cohen 1979 | 242 second and third graders | 21 (8.6%) |
| Giray et al. 1976 | 280 children aged 5 to 18 | 15 (5.6%) |
All three as reported in Crowder’s encyclopedia entry. Nine of Giray’s fifteen were five or six, and one was over ten.
It isn’t an afterimage
An afterimage, from a camera flash for example, moves when your eyes move and usually comes in a different colour. By the criteria Haber lists, an eidetic image stays put while the eyes scan it, lasts longer, and has the original’s colours. It’s also unlike an ordinary mental picture of your bedroom: it fades on its own, part by part, and Searleman says blinking on purpose can destroy it for good.
It isn’t a good record
This is the finding that matters most for the popular idea. Searleman: “the accuracy of many eidetic images is far from perfect”, and people alter details and “even invent some that were never in the original”. The Habers’ own data, Crowder says, showed eidetic imagery is “rather certainly not the agency for good memory of detail”. His conclusion: it “shares practically nothing with the popular concept of photographic memory.” Searleman’s reading is that these images are “reconstructed from memory” and can be swayed by biases and expectations like other memories.
Haber’s 1979 review adds that extensive research failed to find consistent links between eidetic imagery and any cognitive, intellectual, neurological or emotional measure. Crowder notes that Gray and Gummerman (1975) had reservations about how cleanly it separates from very accurate memory. Their paper is one we didn’t read.
How do you test it objectively?
Reports of still seeing a picture are hard to check, so Leask, Haber and Haber (1969) used the fusion method. Two pictures that mean nothing alone appear one after the other, and the person lays their image of the first on the second. Together they make a meaningful picture. Children picked out by the other criteria could do this, Crowder reports, and other children couldn’t.
Why children and not adults?
No one knows, and the sources disagree on more than the summaries suggest. Haber’s review says eidetic images are available to “a small percentage of children 6–12 years old” and are “virtually nonexistent in adults”. Searleman says the same of adults, with “a few notable exceptions”.
The tidy explanation is that language replaces images as children grow up. Searleman offers a version of it: naming things while you look interferes with forming the image, and adults name things more. But Haber’s review casts doubt on the general idea that eidetic imagery is a less mature form of memory. It reports that a long study through the elementary years found the ability “remarkably stable”, and no correlation with abstract thinking or reading.
The age data are less tidy than a simple decline. Crowder calls the fall with age well documented. Haber reports no higher incidence at preschool ages. In Giray’s school-age sample the youngest groups had the highest rate, and beyond age 7 it didn’t vary systematically. Two 1985 reports suggested it rises again among older people, which would make the curve U-shaped. Studies have also looked at children with intellectual disability or brain injury. Crowder says eidetic imagery is probably related to some brain disorders, Haber’s review reports no higher incidence, and this page doesn’t go into them.
Didn’t someone pass a test of exact recall?
One person, as far as we can find. In 1970, Charles Stromeyer and Joseph Psotka reported in Nature on a woman called Elizabeth. As Joshua Foer tells it, Stromeyer showed her right eye a pattern of 10,000 random dots and, a day later, showed her left eye another. She fused them from memory and saw a three-dimensional image, which only someone holding the first pattern could do.
Stromeyer then married her, and she was never tested again. Foer writes that some psychologists have concluded there’s “something fishy” about the findings, and that Stromeyer denies it. When Foer asked him in 2006 he stood by the work: “We don’t have any doubt about our data.” He also said his one-person study “is not strong evidence for other people having photographic memory.” We couldn’t get the Nature paper, so this account is Foer’s.
The obvious next step was tried. In 1979 John Merritt published a test of this kind in magazines and newspapers. By his estimate about a million people tried it. Thirty sent in the right answer, he visited 15 of them at home, and with him watching, none could do it.
The famous claims
| Who | The claim | What we could find |
|---|---|---|
| Shass Pollaks | Talmud scholars said to know which words a pin passed through on each of 5,422 pages | Stratton reported it in 1917. Foer puts it down to a lifetime’s memorising rather than snapshots |
| Kim Peek | said to have memorised 9,000-plus books at 8 to 12 seconds a page | “never been rigorously tested”, says Foer |
| Stephen Wiltshire | the “human camera”, who sketches a scene after looking at it for a few seconds | “He takes liberties”, says Foer |
| S., Luria’s mnemonist | studied for three decades for his memory | seemed to have mastered mnemonic techniques, says Foer |
| Truman Capote | near-total recall of dialogue, so no notes | Foer suspects it was a cover for invented dialogue |
| John von Neumann | could recall every book he’d read, according to Herman Goldstine | a colleague’s account, cited on Wikipedia. We found no test |
So how do memory athletes do it?
They use methods, and it shows in the data. Memory championship competitors have memorised 52 playing cards in 20 seconds and 1,000 binary digits in five minutes, by Müller and colleagues’ account. Asked how, top-ranked athletes credit mainly the method of loci: you imagine a route you know well, leave each item at a landmark along it, then walk the route again to pick them up. Wikipedia adds that none of the world’s best competitive memorizers has claimed a photographic memory.
Practice alone gets you a long way
- Digits. Ericsson, Chase and Faloon (1980) took one subject from a span of 7 digits to 79 in more than 230 hours of practice. The usual figure, about seven, is on the working memory page.
- Pi. Chao Lu set a Guinness record with 67,890 decimals. Tested in the lab, he had a digit span only in the normal range, and his feat came from mnemonic encoding and thousands of hours of memorising.
- A championship. Feng Wang won the World Memory Championship by recalling 300 digits presented one a second, and reproduced a span of 200 in the lab, where Ericsson and colleagues recorded his descriptions of how he encoded them.
- Words. Dresler and colleagues compared 17 athletes with matched controls on 72 words, 20 minutes after studying them: 70.8 recalled against 39.9. Then people with no experience were assigned to six weeks of training or to control groups. The method-of-loci group did 40 sessions of 30 minutes and gained significantly more on the 72 words than both control groups. In the authors’ separate check of progress, they went from 16.6 words on a five-minute task at the start to a mean weekly best of 42.3.
What about their brains?
Here the studies differ. Maguire and colleagues (2003) found superior memory “was not driven by exceptional intellectual ability or structural brain differences” and that the memorisers used a spatial strategy. Ericsson’s commentary on that study is titled “Exceptional memorizers: made, not born”.
Müller and colleagues (2018), with 23 athletes, did find a larger right hippocampus; the left wasn’t significantly different. Without following people over time, they write, they “can only speculate” whether that came before the training or from it. So it’s fair to say that methods do most of the visible work. It isn’t settled whether the athletes began with ordinary brains, though Dresler’s trainees show large effects from training in people with no prior experience.
Chess is the classic example of meaning doing the work
Chess players can seem to photograph a board. In Gong, Ericsson and Moxley (2015), rated players recalled 2.74 chunks of pieces from real game positions and non-players 1.02. With randomly placed pieces the gap fell to 1.30 against 0.98, which wasn’t significant. Their advantage depends on positions that make sense, and a photograph wouldn’t.
How is it different from other kinds of remarkable memory?
| What it is | Lasts | Exact? | |
|---|---|---|---|
| Photographic memory (the claim) | reading back any image or page seen once | claims vary | claimed perfect, unproven |
| Eidetic imagery | a picture reported as still in view, mostly in children | half a minute to several minutes | often not |
| Iconic memory | the brief store everyone has of what they just saw | about a second or less | holds more than you can report, before it fades |
| Trained memory | methods such as the method of loci | gains still measured four months on in Dresler’s study | can be, for what was encoded |
| HSAM | the days of your own life, recalled in detail | much of a lifetime | vivid, and still open to false memories |
HSAM, highly superior autobiographical memory, comes closest in the public mind. The first case was the woman known as AJ, reported in 2006, and Talbot and colleagues’ review says these people don’t use deliberate methods and that memories come to them automatically. It’s about their own lives, a kind of episodic memory.
Patihis and colleagues tested 20 such people against 38 controls on false memory. On related words never shown, the two groups were nearly the same (70.3% and 70.8%). On footage of a plane crash that doesn’t exist, 20% and 29% said they’d seen it, with no significant difference. On misinformation about a slideshow, the HSAM group did worse.
And everyone’s visual memory is better than they think. On the short-term and long-term memory page, ordinary people pick out pictures from thousands they saw once, often with detail. Recognising a picture when it’s in front of you is a much easier job than reading from one that isn’t.
What do people get wrong about it?
That eidetic memory is just the scientific name for it
Wikipedia’s first sentence says eidetic memory is “also known as photographic memory”. Its next paragraph separates them, and Crowder says eidetic imagery shares practically nothing with the popular idea. In practice the words get swapped, so when someone says eidetic memory, ask which they mean: the children’s effect, or the claim about pages.
That an eidetic image is a long afterimage
Foer’s otherwise useful article calls it “essentially a vivid afterimage”, and Wikipedia also says eidetikers report “a vivid afterimage”. Searleman’s account is the reverse: “an eidetic image is not simply a long afterimage”. It holds still when the eyes move and keeps the original colours, and those are among the criteria Haber lists for telling the two apart.
That there’s one figure for how many children
You’ll see 2% to 15% (Foer), about 2% to 10% (Searleman), and two to ten percent of six-to-twelves (Wikipedia). The surveys we could trace, all through Crowder, sit between 5.6% and 8.6% and used similar criteria. Rates vary with the method and the ages tested, and none of these is a measure of perfect recall.
Does an AI have a photographic memory?
A computer does what people imagine: a saved file comes back exact, every time. A language model is a different case. Text in its current context stays available word for word, though it can still misquote it. What it picked up in training isn’t stored as a page it can look up, and it can answer from that with confidence and be wrong.
Counterparts, the memory system whose site this is, keeps what the AI writes down in its own words, so what it stores is a description of what happened. The Field Guide follows mechanisms that AI systems do borrow, and the directory compares the systems.
Questions people ask
What is photographic memory?
The popular idea of being able to look at something once, such as a page, and later see it and read from it as if it were a photograph. It isn't a scientific term, and that ability hasn't been shown in anyone in a test that held up when others tried it.
Is photographic memory real?
Not in the sense people mean. One person seemed to pass a demanding test in 1970 and was never tested again, and a search for others through magazines and newspapers found 30 people who sent the right answer, and none of the 15 visited at home could do it. Trained memories can be extraordinary, but they use methods.
What does eidetic mean?
Eidetic, pronounced eye-DET-ik, comes from the Greek eidos, visible form. Eidetic imagery is a persisting image of a picture that a person reports still seeing after it has gone, for between half a minute and several minutes.
Is eidetic memory the same as photographic memory?
People use them interchangeably, but researchers separate them. Eidetic imagery is a measured phenomenon, mostly in children, and it isn't especially accurate. Photographic memory is the popular claim of perfect recall. Robert Crowder's encyclopedia entry says eidetic imagery shares practically nothing with that idea.
How common is eidetic memory in children?
Surveys using the Habers' criteria found 8% of 150 elementary-school children, 8.6% of 242 second and third graders, and 5.6% of 280 children aged 5 to 18. Alan Searleman gives a range of about 2% to 10% for preadolescents.
Can adults have eidetic memory?
Very rarely, if at all. Ralph Haber's 1979 review calls eidetic images virtually nonexistent in adults, and Searleman says virtually no adults seem to have them, with a few exceptions. Crowder reports two 1985 studies suggesting it rises again in old age.
Is there a photographic memory test?
Not one that can tell you that you have it from a website. The serious version is the fusion method: two meaningless pictures shown one after the other, which make sense only if you can lay your image of the first on the second. The demo on this page is a version of it, and it shows how you did, not what you have.
How do I know if I have a photographic memory?
A web page can't say. The claim is specific: a page you saw once, read back later with the words in place and no errors. The memory athletes and trained subjects in these studies had practised methods.
Can you train yourself to have a photographic memory?
There's no evidence you can. Searleman doubted it and knew of no attempts to teach eidetic imagery. Memory skill can be trained a great deal, though: one subject went from 7 to 79 digits after more than 230 hours of practice.
Which famous people had a photographic memory?
Many are said to have. None of the cases we looked at was shown to be photographic under test. Kim Peek's reading feats were never rigorously tested, according to Joshua Foer, who also says Stephen Wiltshire takes liberties in his drawings and that the mnemonist S. seemed to rely on techniques.
Is highly superior autobiographical memory the same thing?
No. People with HSAM remember the days of their own lives in striking detail, and the ability doesn't appear to come from mnemonic methods. It isn't a picture of a page, though, and in one study they showed false memories much as controls did, and more on a misinformation test.
Questions about how any of this carries over to AI? Ask in r/AIMemorySystems.
References
Sources
- Crowder, R. G. Eidetic imagery. In Learning and Memory (encyclopedia), via Encyclopedia.com. Read in full. Our source for Haber & Haber (1964), Leask et al. (1969), Paivio & Cohen (1979), the numbers in Giray et al. (1976), and the 1985 reports on older adults.
- Haber, R. N. (1979). Twenty years of haunting eidetic imagery: Where's the ghost? Behavioral and Brain Sciences, 2, 583-594. Abstract only; the full paper is paywalled.
- Searleman, A. (2007). Is there such a thing as a photographic memory? Scientific American. Read in full.
- Foer, J. (2006). No one has a photographic memory. Slate. Read in full, with its corrections. Journalism, and our source for Stromeyer & Psotka (1970), Merritt (1979), Stratton (1917), Peek, Wiltshire, Capote and S..
- Giray, Altkin, Vaught & Roodin (1976). The incidence of eidetic imagery as a function of age. Child Development. Abstract only.
- Ericsson, Chase & Faloon (1980). Acquisition of a memory skill. Science, 208, 1181-1182. Abstract only.
- Ericsson (2003). Exceptional memorizers: made, not born. Trends in Cognitive Sciences. Abstract only.
- Maguire, Valentine, Wilding & Kapur (2003). Routes to remembering: the brains behind superior memory. Nature Neuroscience. Abstract only.
- Hu, Ericsson, Yang & Lu (2009). Superior self-paced memorization of digits in spite of a normal digit span. JEP: Learning, Memory, and Cognition. Abstract only.
- Ericsson, Cheng, Pan, Ku, Ge & Hu (2017). Memory skills mediating superior memory in a world-class memorist. Memory. Abstract only.
- Dresler et al. (2017). Mnemonic training reshapes brain networks to support superior memory. Neuron. Read in the parts cited.
- Müller et al. (2018). Hippocampal-caudate nucleus interactions support exceptional memory performance. Brain Structure and Function. Read in the parts cited.
- Wagner et al. (2021). Durable memories and efficient neural coding through mnemonic training using the method of loci. Science Advances. Introduction, for the method.
- Gong, Ericsson & Moxley (2015). Recall of briefly presented chess positions and its relation to chess skill. PLoS ONE. Read in the parts cited.
- Talbot et al. (2025). Highly superior autobiographical memory (HSAM): a systematic review. Neuropsychology Review. Introduction.
- Patihis et al. (2013). False memories in highly superior autobiographical memory individuals. PNAS. Read in the parts cited.
- Wikipedia, Eidetic memory. A secondary source, used for what summaries say and for Goldstine on von Neumann.
Last checked on . Stromeyer & Psotka (1970), Merritt (1979), Stratton (1917), Haber & Haber (1964), Leask et al. (1969), Paivio & Cohen (1979), Gray & Gummerman (1975) and the 1985 reports weren’t read directly, so the notes above say whose description we relied on. Made by Counterparts AI. Counterparts is an open source memory for AI. If something here is wrong, tell us.