Somebody in every friendship group claims a photographic memory. The claim is almost impossible to argue with, because the usual test is a page of text or a busy picture, and both can be passed without anything photographic happening at all. Read a paragraph, remember the gist, reconstruct plausible wording, and you will sound like you recalled the page. Look at a photo of a kitchen, remember there was a kettle, and you can describe a kettle.
Color closes that loophole. There is no word for the exact blue you were shown five seconds ago. English gives you about eleven useful color words and the gamut holds sixteen and a half million codes, so language can only ever hand back a rough neighbourhood. Better still, the error can be measured. Two colors have a defined perceptual distance in CIEDE2000, so a color memory test produces a number rather than an impression.
So I ran the numbers on what a photographic memory would actually have to do. The short version is that ordinary color recall leaves you compatible with a median of 1,555 distinguishable colors, while recall worthy of the word photographic would narrow that to 8. That is a factor of roughly two hundred, and it is the gap that every photographic memory claim has to cross.
Eidetic memory and photographic memory are not the same claim
The two terms get used interchangeably and they should not be. Eidetic imagery is a real, documented phenomenon with a research literature behind it. A child looks at a picture, the picture is taken away, and the child continues to describe it as though still looking at something, scanning it, reporting details in the order they appear rather than in the order they matter. It shows up in something like 2 to 10 percent of children and almost never in adults.
Photographic memory is the stronger and much shakier claim: that somebody can store a page or a scene in full detail and read it back later at will. Ralph Haber, who spent twenty years running eidetic studies, wrote a famously deflating review in 1979 asking where the ghost had got to. His answer was that eidetic imagery is real but far less impressive than the folklore, and that the photographic version had not shown up in the lab at all.
There is exactly one celebrated case that looks like the real thing. Stromeyer and Psotka reported in Nature in 1970 that a Harvard subject known as Elizabeth could hold one half of a random dot stereogram in mind, view the other half the next day, and fuse the two into a depth image. Random dot stereograms are the ideal test precisely because they carry no describable content, so there is nothing to verbalise. The result has never been replicated. Elizabeth declined to be tested again, and the researcher married her, which is not a disqualification but is not nothing either.
One unreplicated case in half a century is a thin foundation. That is roughly where the science sits, and it is worth knowing before anyone takes an online quiz that congratulates them on a rare gift.
Setting an honest bar for photographic
Vague claims survive because nobody defines the threshold. So here is a definition that can be tested: recall is photographic if the color you report back is within one just noticeable difference of the color you were shown. If your memory is accurate to the limit of what your eye could resolve even with both colors side by side, calling it photographic seems fair. Anything looser is a description, not a photograph.
The standard figure for that threshold is 2.3 CIEDE2000, which is where a difference becomes reliably visible under normal viewing. Now the question becomes geometric. If your memory of a color is accurate to within some distance, how many genuinely different colors are still consistent with what you remember?
I measured this directly. For 400 randomly chosen sRGB colors I counted how much of the color cube falls inside a given CIEDE2000 radius, then divided by the size of one just noticeable difference cell measured at that same location at full 8 bit resolution. The second step matters, because sRGB is badly non uniform: a fixed step in code values means very different things in different parts of the gamut, so counting hex codes would be misleading. Dividing by the local cell size cancels that out and leaves a count of actually distinguishable colors.
What the radii translate to
- Within 2.3 CIEDE2000, the photographic bar: a median of 8 distinguishable colors, scoring 9.76 out of 10 on this site.
- Within 5.0: a median of 75, scoring 9.25.
- Within 10.0: a median of 412, scoring 8.08.
- Within 17.9, the level a color name alone gets you: a median of 1,555, scoring 6.30.
That last figure is the one to sit with. 17.9 CIEDE2000 is the expected error if you encode a color using nothing but the eleven basic color words of English, which is the floor measured in our work on aphantasia. A person operating purely on the word blue is carrying enough information to narrow sixteen million codes down to about fifteen hundred distinguishable colors. That is a real achievement for one syllable, and it is nowhere near a photograph.
Note also how quickly the numbers collapse as precision improves. Going from 17.9 to 2.3, a factor of about eight in distance, cuts the candidate set by a factor of nearly two hundred, because the confusion region is a volume and volumes go as the cube. Small gains in accuracy buy enormous gains in specificity. This is exactly why the photographic claim is so demanding, and why it fails so obviously once you put a number on it.
The finding that surprised me: not all colors are equally vague
The median hides something more interesting. Across those 400 reference colors the size of the confusion set at the 17.9 level ranged from 82 to 17,353. Two people could remember two different colors with identical accuracy in CIEDE2000 terms and one of them would have pinned it down two hundred times more tightly than the other.
So I sorted by where the color sits. Sampling 600 colors evenly across hue, saturation and brightness and binning the results gives a clear pattern, and it is not the one I expected.
- Vivid colors, saturation above 67 percent: median 1,125 candidates.
- Medium saturation: median 3,873.
- Muted colors, saturation below 34 percent: median 6,726.
A remembered muted color is about six times more ambiguous than a remembered vivid one, at the same recall accuracy. Brightness barely mattered by comparison: dark, mid and light colors came in at 2,459, 4,519 and 2,274, with no clean trend.
This lines up with something everyone has experienced and nobody has a name for. People are confident about fire engine red and useless about beige. Ask someone to pick their living room wall colour off a paint chart and watch the confidence evaporate, then ask them the colour of a stop sign. The difference is not that beige is boring. It is that the region of color space compatible with a remembered beige is genuinely enormous, so the same quality of memory delivers a far worse answer. Greige, taupe, ecru and mushroom are not marketing nonsense, they are a vocabulary invented under pressure because ordinary memory cannot separate those colors at all.
It also means any color memory test that draws mostly saturated colors will flatter its players, and one that draws pastels will look brutally hard, without either test being more valid. Worth knowing if you build these things.
What the memory research says is actually going on
The laboratory picture of color memory is well developed and it does not involve photographs anywhere.
Zhang and Luck ran the definitive version of this experiment in 2008, showing people colored squares and asking them to report a color back on a color wheel. Their result was that visual working memory holds a small number of items, roughly three or four, at a fixed resolution. Showing six items instead of three did not make the remembered colors blurrier. It made people more likely to have stored nothing at all about a given item. Memory behaved like a small number of slots rather than a photograph that gets progressively grainier.
That single word, slots, kills the photographic metaphor on its own. A photograph degrades smoothly and covers the whole frame. Human visual memory covers three or four things well and the rest not at all.
Bae and colleagues added the second mechanism in 2015. Remembered colors do not just lose precision, they drift, and they drift toward the best example of whichever category the color fell into. A slightly purple blue is remembered as more prototypically blue than it was. Memory is quietly mixing a continuous estimate with a category label, which is efficient and also means the error has a direction rather than being random noise. We have written separately about how this shows up in memory colors, where familiar objects get remembered as more saturated versions of themselves.
There is one place where something photographic genuinely exists. Sperling demonstrated in 1960 that immediately after a display vanishes, people hold a high capacity and essentially complete visual trace, and can report any part of it if cued fast enough. That is iconic memory. It is the real photograph, and it lasts a fraction of a second before it is gone. Everyone has photographic memory. It expires in under a second and cannot be searched afterwards.
Our view: the question is badly posed
After running these measurements the honest conclusion is that photographic memory is not so much false as incoherently specified. When somebody says they have one, they usually mean something real: unusually good recall for layout, for where on the page a fact appeared, for the shape of a diagram. Visual long term memory for object detail is genuinely large, as Brady and colleagues showed in 2008 when people recognised thousands of images with fine grained accuracy after a single viewing. That is impressive and it is not photographic, because recognition is a far easier task than reproduction.
The gap between recognising and reproducing is where the whole myth lives. You can recognise the exact blue of a brand logo instantly when it is put in front of you, and you cannot pick it off a color wheel to within eight candidates. Both facts are true about the same memory. Tests built on recognition make people look photographic, and tests built on reproduction do not, which is why the format of an online memory quiz predicts its result better than the person taking it does.
If you want to know where you actually stand, use a test that forces reproduction on a dimension words cannot carry. That is what our color memory game does: it shows a color, takes it away, and asks you to rebuild it. The scoring is CIEDE2000 underneath, so the number you get out is the same currency as everything above. If you are consistently scoring above 9.7 you are recalling colors to within a just noticeable difference, and you have a legitimate claim to something the literature would find interesting. Most good players sit well below that, and so do I.
How to run the test properly on yourself
- Test reproduction, not recognition. Being shown four options and picking one is a different and much easier task.
- Use colors words cannot pin down. Muted and unusual colors are the honest ones, though they will feel unfair for the reason measured above.
- Watch for the drift, not just the size of the error. If your misses land consistently toward the textbook version of the colour, that is the categorical effect and not a failure of storage.
- Take the score across many rounds. A single lucky round tells you nothing, which is the main flaw of most quizzes built on one image.
The related tests on this site probe adjacent abilities if you want a fuller picture. The color memory test covers what the score means, and how long you can remember a color covers the decay side, which is the other half of the story and the reason eidetic imagery in children fades within minutes rather than being filed away.
The short answer
Eidetic imagery is real, mostly found in children, and much weaker than the folklore suggests. Photographic memory in the sense people mean it rests on a single unreplicated case from 1970. Tested on color, where words cannot rescue you and the error can be measured, ordinary human recall lands about two hundred times short of the photographic threshold, and lands six times further short again if the color happens to be muted. The remarkable thing is not that people fail this test. It is that a single word like blue still narrows sixteen million possibilities down to about fifteen hundred.