Free, 108 color choices, scored against a published cutoff

Synesthesia test scored on whether you repeat yourself

Pick the color of each of 36 characters, three separate times, with the order reshuffled every round and your earlier picks hidden. The score is the mean distance between your three choices for the same character, compared against the 1 RGB distance threshold from Eagleman, Kagan, Nelson, Sagaram & Sarma (2007) — the one figure in this area that was defined on an actual measurement rather than on how vivid somebody said their letters felt. Free, no account, and each round is kept in your own browser so you can leave a week between them, which is exactly what makes the result worth having.

100% free
No signup
36 characters × 3 rounds
Resumable across days
Cutoff 1 RGB distance

Three passes over the same characters, in three different orders

Pick the color each letter is for you. Then do it again, and once more, with the characters coming in a different order every time and your earlier choices hidden. The score is how far apart your three choices for the same character are — nothing about how strong or how obvious the color felt, because consistency is the part that can be checked and vividness is not.

Leave as long as you can between rounds. Three passes in one sitting mostly measure whether you remember what you picked ten minutes ago, and the published cutoff was not defined on that. Each finished round is saved in this browser so you can close the tab and come back tomorrow; the results panel prints the real gap between your rounds beside your score, and says what a short gap does to it.

Starting round 1 discards anything saved here from an earlier attempt. A grapheme that has no color for you gets a “no color” button rather than a forced guess — a real mapping is usually patchy, and pages that force a color on all 36 manufacture consistency out of indifference.

How to run the three rounds properly

The gap between rounds is not an inconvenience; it is the measurement.

Set a color for every character in round 1
Each character appears on its own and you set a color on three labeled axes — hue, saturation, lightness — with the hex code and the nearest color name shown as you move them. A character that genuinely has no color for you gets a no-color button instead of a forced guess; real mappings are patchy, and a page that makes you color all 36 manufactures agreement out of indifference. There is no going back to a character you have passed, and no clock on any of it.
Close the tab and come back another day
The finished round is written to this browser and nothing else. Leave as long as you can stand — a day is enough, a week is better — because three passes in one sitting largely measure whether you remember what you picked twenty minutes ago, and the threshold this page compares you against was measured on people who had genuinely forgotten. If you would rather not rely on the browser, the round-end screen gives you the whole session as copyable text to paste back later.
Read the spread, the gaps and the per-character table
After the third round you get the mean spread across all characters that were colored in all three rounds, the real elapsed time between each pair of rounds, and a table showing your three picks for every character with its own spread. Characters you marked colorless throughout are excluded rather than scored as perfectly consistent, and characters colored in some rounds and not others are counted separately, because that is an inconsistency this metric cannot express as a distance.

Technical specifications

Characters per round	36 by default — A to Z plus 0 to 9, the set the published battery runs — or 26 letters only, which is 78 choices instead of 108. Digits are usually colored by the same mapping as letters, so the shorter set costs less than it looks
Rounds and order	3, each shuffled independently from one session seed, so no character falls in the same position twice and the seed is printed with your result. Your earlier picks are hidden until all three rounds are in — showing them would turn round 2 into a copying exercise
Color input	Hue 0-359, saturation 0-100, lightness 0-100, on three keyboard-operable sliders with a live hex code and the nearest CSS color name. The value stays continuous, which is what keeps the score on the axis the cutoff is defined on; a fixed palette would make an exact repeat trivially available and quietly lower everyone's score
Distance metric	Euclidean distance in raw RGB with each channel on 0-1. A character's spread is the sum of the three pairwise distances between its three choices, so it runs 0 to about 5.2 in principle and rarely past 2 in practice; the headline is the mean of those spreads
Threshold	1 RGB distance from Eagleman, Kagan, Nelson, Sagaram & Sarma (2007). Deliberately measured in raw RGB rather than a perceptually uniform space, because a figure computed in CIELUV is not on the same axis as this cutoff and would need the threshold from Rothen, Seth, Witzel & Ward (2013) instead
Refusals and partials	A character marked colorless in all 3 rounds is excluded from the mean and reported separately. One colored in some rounds and not others is counted as a partial and also excluded, because there is no defensible distance between a color and an absence
Gap between rounds	Recorded to the millisecond and reported in plain words. Any gap under 10 minutes is called out, and a full set of short gaps makes the results panel report the score as a ceiling on your consistency rather than a measurement of it
Where the run lives	One local-storage entry in this browser, holding your seed, your set, your picks and the finish time of each round. Nothing is sent anywhere, no reaction time exists on this page, and the delete button removes the entry outright

Frequently asked questions

Why does the test not ask what color each letter feels like?

Because that question cannot separate synesthesia from an ordinary learned association, and every version of this test that asks it is really running a preference survey. Anybody who owned a set of colored fridge magnets will produce confident answers, and will produce the same confident answers ten minutes later because they remember giving them. What the research uses instead is repetition without recall: the same character presented again after enough time that the last answer is gone. A synesthete's mapping holds; an association drifts. Nothing about how strong, obvious or emotionally loaded the color felt is asked here, because none of that can be checked from outside.

What does the number actually mean, and what would a good score be?

It is the average distance in color space between your three choices for the same character, so lower means more consistent and zero would mean identical picks every time. 1 RGB distance is the line Eagleman, Kagan, Nelson, Sagaram & Sarma (2007) used to keep a participant inside a synesthete sample, and being under it means your repeats landed closer together than that threshold on the run you just did. It is not a probability, not a percentile, and not a verdict: no page can convert one browser run into odds that you are a synesthete, and any page offering you a percentage is inventing it.

Can I take all three rounds in one sitting?

You can, and the page will tell you what it cost. Working memory holds a color choice for far longer than twenty minutes, so three rounds back to back partly measure recall — which pushes the spread down and makes almost anybody look consistent. That is why the results panel prints the real elapsed gap between each pair of rounds, flags any gap under ten minutes, and describes a fully short-gapped score as a ceiling rather than a measurement. The honest way to use the page is to do it twice: once in a sitting, once with days in between, and compare the two figures. The difference between them is more informative than either alone.

Some letters have no color for me. Does that break the test?

No, and being able to say so is important to getting the number right. Real grapheme-color mappings are commonly patchy: vowels and early letters are often strongly colored while several consonants are simply not, and forcing a choice on those turns indifference into data. Each character therefore has a no-color button. A character you mark colorless in all three rounds is left out of the mean rather than counted as a perfect repeat, which would otherwise reward not knowing. A character you color in some rounds and not others is reported on its own, because there is no meaningful distance between a color and an absence.

I see colors for days of the week and for music, not for letters. Is that here?

Not as a test, and it is worth being straight about why. Grapheme-color is only one variant and not the most common one — colored weekdays and months are more frequent — and the others each need their own instrument: an ordinal sequence has no fixed inventory to shuffle, sound-color needs stimuli this site does not play, and spatial forms are drawn rather than picked. A consistency test only works when the same discrete item can be re-presented without the visitor recognizing the trial, and days of the week give you seven items, which is not enough to average over. So this page covers the variant it can measure properly and names the rest instead of pretending to test them.

Would color blindness change my result?

It changes what the picker can capture, not whether you have the trait, and the page is built around that. Two colors that are far apart in RGB can be indistinguishable to someone with a red-green deficiency, so three picks that look identical to them can score as inconsistent — the metric would then be measuring their retina. That is why the choice is made on labeled numeric axes with the hex code and the nearest color name shown, rather than by hunting on a color wheel: the pick can be set and verified by number and by word. It is also worth knowing that synesthetic colors are not constrained by what a person can see; some synesthetes describe colors they have no name for and cannot point at.

Why measure distance in RGB when it is not perceptually uniform?

Because the cutoff on this page is defined in RGB, and mixing the two would produce a comparison between numbers on different axes. RGB distance genuinely overstates some differences and understates others — the same numeric gap is much more visible in green than in blue — and Rothen, Seth, Witzel & Ward (2013) refined the method into a perceptually uniform space for exactly that reason. Switching this page to that space would require that paper's threshold rather than 1 RGB distance, and would make your figure incomparable with every published score from the original battery. Keeping the metric crude and matched to its threshold is the more honest of the two options.

Consistency, and why it is the only part of this that can be checked

Synesthesia has been described since the nineteenth century and doubted for most of that time, and the doubt was reasonable: everything anybody knew about it came from people reporting their own experience, and reports are cheap. What settled the question was not a better interview but a measurement. If the letter A really pulls a specific orange, that orange should still be there when the memory of saying so is gone — and if it is an association picked up from a toy or a poster, it should drift. A standardized battery built on that idea arrived in 2007 and turned a century of anecdote into something with a threshold on it. Full citation, because this figure gets quoted loose: Eagleman, Kagan, Nelson, Sagaram & Sarma (2007), A standardized test battery for the study of synesthesia, Journal of Neuroscience Methods. Later refinement: Rothen, Seth, Witzel & Ward (2013), Diagnosing synaesthesia with online colour pickers: maximising sensitivity and specificity, Journal of Neuroscience Methods.

Three details in that design are what most online versions drop, and dropping them is what makes their numbers meaningless. The first is the delay: repeats have to be far enough apart that recall is not doing the work, which is why this page saves each round and asks you to come back rather than running all three in six minutes. The second is that earlier choices must be hidden — a page that shows you last round’s swatch is testing whether you can match a swatch. The third is that a colorless character has to be allowed to be colorless. Real mappings are uneven, and a version that demands a color for all thirty-six characters gets its consistency partly from people who shrugged and dragged the slider to roughly where they left it. The prevalence figure is worth the same care: 1-2% for this variant, from Simner et al. (2006), Synaesthesia: the prevalence of atypical cross-modal experiences, Perception, and the reason it is a range rather than a decimal is that every study drew the verification line in a slightly different place.

What the score cannot do is tell you which side of a category you are on, and this is where the page parts company with most of its search results. The distribution of consistency is continuous — there is no gap in it — so the threshold is a decision made for the purpose of assembling a clean study sample, and living just under it or just over it is a fact about one run rather than about a person. The other trait pages on this site take the same view of their own numbers: the aphantasia test pairs a questionnaire with a task precisely because the questionnaire alone is a self-report, and the face blindness test reports percent correct with an interval instead of a norm it cannot defend. If you came here from a list of curiosities, the misophonia test and the sensory processing test cover sound and sensory sensitivity, the alexithymia test covers naming feelings, the time blindness test covers estimating intervals, and the left or right brain test and the VARK learning style inventory are the two pages here that report the evidence against the idea they are named for. There is no timing anywhere on this page, so none of the millisecond caveats that govern the speeded tests on this site — the Mensa-format practice test among them — apply to any figure here.

This is a measurement exercise, not a clinical assessment. It reports what you did on this page against a stated reference and nothing more — it cannot establish grapheme-color synesthesia. Only a qualified professional, working with more than a browser, can make that judgment.

Where your 108 color choices are kept

Every number on this page is worked out by JavaScript running in the tab you are reading it in. Your answers, your reaction times and your score are never uploaded, logged or kept — which is also why the test carries on working after you disconnect from the network, and why nothing here can be held back behind an email address.

This is the one page on the site that writes something down, and it is worth being exact about what. One local-storage entry in this browser holds your seed, which character set you chose, the color you gave each character in each round, and the moment each round finished — because a test that has to span days cannot work otherwise. It is readable only by this site in this browser, it is never transmitted, and the delete button on the results panel removes it. The copyable session text is the same data, handed to you.