Color Blindness Test

How to take it, and how to read the result

1. Fix the viewing conditions first. Night mode off, OS color filters off, brightness normal, screen straight on, no colored lamp or sunset light on the panel. A phone in auto-brightness at 20% will fail plates a healthy eye should pass.
2. Answer fast, but do not guess. A real plate is either obvious or it isn't. If you have to hunt for the shape, that is already information — press Nothing / can't tell rather than picking a digit that might be right.
3. Watch the control plates. Three of the sixteen are built so that every deficiency type still sees them clearly. If you miss those, the problem is almost certainly the screen or the lighting, and the rest of the score means nothing.
4. Read the breakdown, not the total. The score is split into the protan line (red-weak), the deutan line (green-weak) and the tritan line (blue-yellow). Missing six red-green plates and zero blue-yellow ones is a pattern. Missing two of each is usually noise.
5. Take it twice, in different light. If the same plates fail both times, that is a consistent finding worth taking to an optometrist. If a different set fails each time, you were guessing.

Before you trust this result

👁 Color blindness simulator

What these plates actually measure

Normal color vision runs on three cone types: L (long wavelength, loosely "red"), M (medium, "green") and S (short, "blue"). Any color you can see is a triple of cone responses. Take one cone away and that three-dimensional space collapses to two dimensions — a whole family of colors that used to be distinct now lands on the same point. The set of colors that collapse together is a confusion line, and the entire Ishihara principle is to paint a digit in one color and its surround in another color on the same confusion line.

That is exactly how the plates on this page are built, and it is arithmetic rather than art. The page converts sRGB into cone excitations (L, M and S) using the standard Hunt-Pointer-Estevez transform, then finds the direction in color space that moves exactly one cone and leaves the other two untouched. Slide along the L-only direction and an eye without L cones registers no change in the arithmetic at all, because the two colors produce identical photoreceptor signals. The one thing left over is the rounding to 8-bit sRGB on the way out: measured on the rendered plates, that leaves a residual of about 1.5% in the channels the deficient eye still has, against a 17–20% signal for everybody else. It is randomly signed dot by dot and uncorrelated with the figure, so it is a noise floor rather than a clue — but it is not literally zero, and on a panel that is not sRGB it will be larger. So each plate picks a mid-gray base, spreads the ground dots along one segment of that direction and the figure dots along another. Typical vision sees two obviously different colors. The eye missing that cone sees one flat field.

Two consequences are worth stating plainly. First, this is derived from the cone fundamentals, not from the simulation matrices used by the simulator above — those are a rendering model, and where the two disagree slightly you may see a faint residual in the simulated view of a plate. The cone construction is the stricter one. Second, there is no such thing as a plate hidden from red-green deficiency in general: protan-invisible is the L axis, deutan-invisible is the M axis, and the only color in both is no color at all. What actually happens is that a protan-line plate still leaves a usable signal for a deuteranope and vice versa — which is precisely what lets the score lean one way or the other.

Why brightness cannot give it away

If the figure were even slightly brighter than the ground, anybody could read the plate by luminance alone and the test would measure nothing. Two things prevent that. First, brightness is itself built out of cone responses — so if the two remaining cones report identical values, there is no brightness difference for that viewer either, not just no hue difference. Second, every dot's color is multiplied by a random factor between 0.78 and 1.22 before it is drawn, so brightness varies dot to dot far more than any systematic difference between the two groups could. The dot's radius is drawn before the code even asks whether the dot falls inside the digit, so size cannot leak the shape either, and the figure and ground each span a range of colors rather than being one flat hue — the variation you can see is real, and it carries nothing.

Anomalous trichromacy vs dichromacy

Most people with a color vision deficiency are anomalous trichromats: all three cone types are present, but one has a shifted absorption peak, so the L and M responses overlap more than they should. They see color, just with less discrimination along one axis — mild cases pass many plates, severe cases fail almost all of them. Dichromats (protanopia, deuteranopia, tritanopia) are missing a functional cone type entirely, and the confusion lines are absolute. This test is calibrated toward dichromacy, which is exactly why a mild anomalous trichromat can score well and still find real-world color tasks hard.

What a bad result looks like, and what to do

Consistent misses on the red-green plates with clean control plates is the finding that means something. Take a screenshot of the breakdown, then book an optometrist — an Ishihara set on paper takes them two minutes, and if it is positive they can run an anomaloscope or a Farnsworth D-15 to say which type and how severe.

Missed control plates means stop and fix the display. The usual culprits: a blue-light filter still running, brightness at 15% on a laptop, a color profile that was never calibrated after a GPU driver update, an OS accessibility color filter left enabled, or a browser rendering into a wide-gamut space without conversion. Re-run the test once the control plates are trivially readable.

Everything missed, including controls almost always means the page is not showing color at all — grayscale accessibility mode, a monochrome projector, or a very aggressive contrast setting. Check the simulator tiles above: if all four look the same, the browser is not painting the colors you think it is.

Tritan-only misses deserve special skepticism. Inherited tritanopia is put at fewer than 1 in 30,000, and blue-yellow loss is more often acquired — from glaucoma, diabetes, ageing lenses or certain medications — than inherited. Screen blue is also the most fragile channel there is — filters, panel age, LED backlight spectra and color profiles all attack it first. Rule out the display before believing the plates.

Who has it, and what it actually means

Roughly 1 in 12 men and 1 in 200 women of Northern European descent have some form of color vision deficiency. The lopsided ratio is not a coincidence: the genes for the L and M cone photopigments sit on the X chromosome, and the condition is recessive. Men have one X, so a single faulty copy is enough. Women have two, so a working copy on the other X usually compensates — which also means women are far more often carriers than affected. Blue-yellow (tritan) deficiency sits on chromosome 7 instead, is not sex-linked, and is much rarer.

"Color blind" is a bad name for it. Almost nobody with the common forms sees in grayscale — that is achromatopsia, which affects about 1 in 30,000 people and usually comes with poor acuity and severe light sensitivity. The everyday reality of red-green deficiency is a full color world where certain pairs collapse: red and green traffic-light LEDs by color alone, ripe and unripe fruit, red text on a green background, the red and green status dots in half the software ever written, blood against skin, and heather on a hillside. Plenty of people reach their thirties without knowing, because nobody ever asks you to name a color you have never seen named differently.

Jobs that still screen for it: commercial and military pilots, air traffic control, train drivers and much of the rail industry, merchant marine deck officers, many police and fire services, some armed-forces trades, and roles involving electrical wiring color codes or histology and pathology staining. Requirements vary a lot by country and by role — a mild deutan may pass a lantern test that a strong protan cannot. If a career depends on it, get tested properly rather than trusting any screen, including this one.

On the design side, the fix is boring and effective: never let color be the only channel. Add a shape, an icon, a label, a pattern or a position. If your chart has a red line and a green line, dash one of them. If your status pill is red or green, put a or a inside it. The simulator above will tell you in five seconds whether your palette survives.

Related tools