Synesthesia is a condition where stimulating one sense, or one kind of thought, reliably produces an experience in another. A letter arrives with a color. A piano note arrives with a color. A weekday has a position in space, or a word has a taste. The person is not imagining it on purpose and cannot switch it off, and the pairings stay the same for years.
The most studied form is grapheme-color synesthesia, where letters and digits carry colors. For a lot of people with it, A has always been red and nobody else seemed to notice. Sound-to-color synesthesia, often called chromesthesia, is the next most common in research samples.
This piece covers what synesthesia is and how common it is, but most of it is about one specific thing we are well placed to measure. The way researchers confirm that someone really has grapheme-color synesthesia is a color test. You pick the color of each letter three times, and if your picks land close enough together, you pass. That is a color memory test, and color memory is the one thing this whole site is built to score. So we took the published pass line and converted it into the same CIEDE2000 units our games use.
The short version: the cutoff is a lot more forgiving than it sounds, and it is looser than the precision you get from remembering only a color's name.
How common is synesthesia
The figure you will see most often comes from Julia Simner and colleagues in 2006, who screened people without letting them self-select, then checked claims with objective tests. They found synesthesia of some kind in roughly 4 percent of people, and grapheme-color synesthesia in about 1 percent. The older estimates of 1 in 2,000 or worse came from waiting for synesthetes to come forward, which mostly measured who was willing to talk about it.
A larger 2015 study by Carmichael and colleagues tested 2,847 people with the online Synesthesia Battery. 4.9 percent said they saw colored letters or digits, but only 1.2 percent passed the consistency cutoff. That gap between saying it and passing it is what the rest of this article is about.
How the synesthesia color test works
The standard tool is the Synesthesia Battery, published by David Eagleman and colleagues in 2007 and still online. The grapheme test shows you each letter and digit three times in random order, and each time you pick its color from a full 16.7 million color picker. There is also a "no color" button.
Your score for a letter is the distance between your first and second pick, plus first and third, plus second and third. Picture the three picks as the corners of a triangle; the score is its perimeter. Average that over every letter and you have your consistency score. Lower is more consistent.
Two cutoffs are in common use:
- Eagleman 2007, RGB. Distances measured in RGB with each channel scaled from 0 to 1. A mean score below 1.0 counts as synesthetic. Rothen and colleagues later found 1.43 worked better for RGB when the distance is measured channel by channel (city block).
- Rothen and colleagues 2013, CIELUV. Distances measured in CIELUV, a color space designed so that equal distances look roughly equally different. Using people who did and did not report synesthesia, they found 135.3 was the cutoff that best separated the two groups. Most current studies use this one.
Rothen's move to a perceptual color space was the right call. RGB distances are notoriously uneven, as our own guide to reading hex codes shows. But CIELUV is from 1976 and it is not uniform either, which is why the color science world moved on to CIEDE2000. So what does 135.3 actually look like?
What the cutoff looks like in CIEDE2000
A perimeter of 135.3 split evenly across three sides is 45.1 CIELUV units per pair. That is the loosest a perfectly evenly spread synesthete can be and still pass.
To convert it, we sampled the sRGB cube on a 52 by 52 by 52 grid, 140,608 colors, and collected 35,201 pairs of colors sitting 45.1 CIELUV units apart (within 3 percent). Then we measured each pair in CIEDE2000 under a D65 white point, the same formula the main color memory game uses to score you.
- Median: 21.75 CIEDE2000 between picks at the cutoff
- Middle 80 percent: 13.17 to 34.24
- Extremes: 5.25 to 47.73, a 9x spread for the same CIELUV distance
- In noticeable differences: 9.5 JNDs, using the 2.3 unit threshold from our just noticeable difference measurement
Two pairs from the sample make the unevenness concrete. Both sit 45 CIELUV units apart.
5.21 CIEDE2000. Two blues most people would call the same blue.
47.0 CIEDE2000. A dark olive and a dark burgundy.
The pattern holds across the gamut. For pairs anchored on dark colors (L* under 30) the median CIEDE2000 at the cutoff is 28.09; for light colors (L* over 60) it is 19.94. CIELUV underrates differences between dark colors, so the test is stricter for someone whose letters are dark and more lenient for someone whose letters are bright and pale. Nobody chooses their synesthetic colors, so this is a small unfairness built into the diagnosis.
The RGB cutoffs land in the same place. Eagleman's 1.0, split across three pairs as a straight-line RGB distance, comes to a median of 20.48 CIEDE2000. The 1.43 city block cutoff comes to 20.43. Three different scoring systems, all of them letting picks drift around 20 CIEDE2000 apart.
One letter, three picks, still a pass
Here is a triangle of picks at the cutoff, built around a typical red. Treat each one as the answer to "what color is A" on a different trial.
#E02129 · the anchor red
#AA2D14 · 11.3 CIEDE2000 from pick 1
#FA4B87 · 21.0 from pick 1, 30.2 from pick 2
A brick red and a hot pink, for the same letter, and the letter passes. The test is not checking whether you saw the same color three times. It is checking whether you stayed in roughly the same region.
The cutoff is looser than remembering a color by name
This is the comparison that made us write this piece. In our aphantasia and color memory article we measured how much color a word carries. If all you keep of a color is that it was "green", and you then produce a green, the expected gap between what you saw and what you made is 17.9 CIEDE2000. That is the floor for verbal color memory, using only the eleven basic English color terms.
At the cutoff, a synesthete's picks sit 21.75 apart. On this site's ten point scale, that is a score of about 5.6 per round. Someone remembering only the word scores about 6.3.
So we ran the obvious simulation. Imagine someone with no synesthesia at all who decides, before the test, that each letter has a basic color word. A is red, B is blue, and so on. On every trial they pick any color at all from inside that word's region of the color space, at random. That is a worst case for them, since real people drift toward the typical shade.
We split the sRGB grid into eleven regions by nearest basic color term, using CIEDE2000 to these approximate focal colors: black #000000, white #FFFFFF, gray #808080, red #E0202A, orange #F28C1E, yellow #FFE11A, green #2E9E40, blue #2255CC, purple #7B2D9B, pink #F4A6C0 and brown #7B4A26. Then we drew 20,000 random triples from each region and scored them the way the Battery does.
Share of triples under the 135.3 cutoff, and the mean perimeter, by color word:
Then we simulated 4,000 whole test takers: 26 letters, each given a basic color word, three random picks per letter, score averaged across the alphabet the way the Battery does it.
- Words chosen evenly across the eleven terms: median score 129.3, 72.4 percent pass
- Words weighted by how much of the gamut they cover, which piles more letters onto the wide green and purple regions: median 139.7, 32.7 percent pass
- Only the eight chromatic words, no black, white or gray: median 134.6, 52.8 percent pass
So a person picking completely at random inside a color word sits right on the line. Now make them slightly more human. Restrict each pick to the half of the region closest to the typical shade, which is still a wide spread. Every version of the simulation then passes 100 percent of the time, with median scores between 98.8 and 106.5.
For comparison, someone clicking anywhere in the color picker at random scores 316.1. The test is very good at catching that person. It cannot catch someone who has decided what color each letter is and sticks to it.
What that does and does not mean
It does not mean the research is wrong or that synesthetes are faking. In practice, non-synesthetes who take the test are not running a deliberate word strategy. They are told to pick whatever color comes to mind, they guess, and they end up inconsistent. That is why 135.3 separates the groups well in real data. It is also why researchers pair the test with self-report and treat consistency as one criterion, not the definition.
What it does mean, in our view, is that the consistency test measures stability of an association, not the presence of a color experience. Those are different things. A learned rule is stable. So is a real synesthetic color. The test cannot tell them apart, and a cutoff of about 9.5 JNDs per pair is wide enough that a well rehearsed verbal rule will clear it.
Two findings from the research back this up from the other direction. Witthoft and Winawer reported in 2013 that ten of eleven synesthetes they studied had letter colors closely matching a Fisher-Price magnet alphabet sold in the 1970s and 1980s, and a 2015 follow-up with Eagleman found the same magnet colors in about 6 percent of 6,588 synesthetes. A toy with a small fixed palette is about the most stable association source there is. Then in 2025 Root and colleagues showed the best cutoff shifts between languages: Spanish speakers needed a significantly looser line than 135.3 and German speakers a stricter one. A number that moves with language is a population statistic, not a fixed property of synesthesia.
Our suggestion, for what it is worth from a color game site: report consistency in CIEDE2000 rather than CIELUV or RGB, so the dark-versus-light unfairness goes away, and treat a pass as "has stable color associations" rather than "is a synesthete". The tools to do that are free and the conversion is a few lines of code.
How few colors the test can tell apart
One more number fell out of the same grid. If two picks need to be about 45 CIELUV units apart before the test treats them as different answers, how many different answers fit in the whole sRGB gamut? A greedy packing of the 140,608 grid colors, always adding the color farthest from every color already chosen, stops at 28 before the next one would land closer than 45.1.
That is barely more than the 26 letters of the alphabet. A greedy packing is a rough estimate, not a hard maximum, but it gives a sense of scale: at the resolution of the diagnostic test, there is room for about one distinct color per letter and almost nothing else. Our how many colors can you see piece puts the number of colors a person can tell apart on a screen in the hundreds of thousands. The test works at a much coarser grain than the eye does.
Chromesthesia and other types
Chromesthesia, where sounds or music produce colors, is tested the same way with tones or chords instead of letters. Researchers have found the best cutoff for it is higher than for letters, which fits the analysis above: musical colors seem to be naturally broader and more variable, so a test tuned on letters would reject real sound-color synesthetes. There are dozens of other documented types, including colored weekdays and months, numbers with personalities, and sequences laid out in space. Many people have more than one.
Synesthetic associations are not random across people, either. Simner and colleagues found in 2005 that A is red more often than chance would allow, and that people without synesthesia lean the same way when asked to pick, which points to shared learning through letter names, frequency and early reading materials. The Stroop effect is the other classic tool here. Synesthetes are slower to name the ink color of a letter printed in the "wrong" color for them, the same interference everyone shows with color words.
So do you have synesthesia
If you have always seen letters or numbers in specific colors, you might, and the Synesthesia Battery is free to take. Go in knowing that a pass shows stable associations and a fail does not rule anything out. Plenty of people who report clear colored letters fail the cutoff, and the 2025 study above found some of them still showed the Stroop interference that marks a real automatic color.
What the Battery does not test is how precise your color memory is. That is a different skill, and it is the one our games measure. The color memory game shows you a color, hides it, and asks you to rebuild it, scored in the same CIEDE2000 used above. The daily challenge gives everyone the same five colors so you can compare. If you want to see how much a color word really holds, the name that color mode is the closest thing we have to the verbal strategy in the simulation, and our piece on color psychology covers the separate question of what colors are supposed to mean.
A synesthete at the diagnostic cutoff would average around 5.6 out of 10 per round on our scale. We would be curious to know whether people with strong grapheme colors do better than that at remembering colors that are not attached to letters. As far as we can find, nobody has tested it.