The Ishihara Test
A field of colored dots with a number hidden inside it, published in 1917 and still the first screening anyone reaches for. Here is what it is actually doing, and where it stops being useful.
Where It Came From
Shinobu Ishihara, a military physician and later professor at the University of Tokyo, was asked to build a color vision screen for army conscripts. He published the first series in 1917, hand-painting the plates in watercolor and using the Japanese hiragana syllabary before switching to Arabic numerals for the international editions.
More than a century later it is still the standard first-line screen, printed in editions of 14, 24 and 38 plates. It survived because the underlying idea is very good and does not depend on any instrument: it hides the figure along an axis the deficient eye cannot see, so no amount of effort or attention recovers it.
The Trick That Makes It Work
Two things separate the figure from the background, and the whole design turns on which one is doing the work.
A plate could set the figure apart by brightness, in which case anyone can read it, color vision or not. Ishihara's plates instead set it apart by hue alone, choosing figure and background colors that sit on the same confusion line, the set of colors that a particular deficiency collapses into a single color.
The random dot mosaic is not decoration. It has two jobs: it breaks up the edge of the figure so no shape cue survives, and it varies dot size and lightness so that brightness carries no usable signal either. That leaves hue as the only route to the answer, which is precisely the point.
The Six Classes of Plate
The full printed series is not fourteen copies of the same idea. Each class does a different job:
| Class | What happens | On this site |
|---|---|---|
|
Demonstration
Confirms the person understands the task and can see the plate at all. Tests nothing about color.
|
Readable by everyone, including a total monochromat, because the figure differs from the background in brightness as well as hue. | Yes |
|
Transformation
The strongest single piece of evidence in the series, because a specific wrong answer is much harder to produce by chance than a blank.
|
A trichromat reads one number; someone with a red-green deficiency reads a different one. | Not yet |
|
Vanishing
The workhorse of the test, and the class most people picture when they think of Ishihara plates.
|
A trichromat reads a number; a deficient observer sees only dots. | Yes |
|
Hidden digit
Catches malingering, since someone faking deficiency does not know what they are supposed to see.
|
The reverse: a trichromat sees nothing, while a deficient observer reads a number. | Not yet |
|
Diagnostic (classification)
Tells protan from deutan, which the vanishing plates alone do only weakly.
|
Two figures side by side, one on a protan confusion line and one on a deutan line. Which half you read separates the two. | Not yet |
|
Tracing
For children and anyone who cannot read numerals.
|
Winding colored lines to follow with a finger rather than digits to name. | Not yet |
How to Take One Properly
The printed test has conditions attached, and they are not optional formalities, the plates were designed under them.
- Daylight, or a daylight-balanced lamp. Tungsten light shifts everything towards red and produces false passes on the protan plates.
- Held at about 75 cm, roughly perpendicular to the line of sight.
- Three seconds per plate. Longer, and people start reasoning their way to an answer instead of reporting what they see.
- Usual glasses or contacts on, but nothing tinted.
- One eye at a time, if the question is whether something has been acquired rather than inherited.
The three-second rule is the one people break most, and it is the one that matters most. A deficient observer given thirty seconds can sometimes assemble a figure from dot-size accidents and edge artefacts. That is a puzzle being solved, not a color being seen, and it produces a pass that means nothing.
What It Cannot Tell You
The Ishihara series is a screen. It answers "is there probably a red-green deficiency here?" very well, and it is a poor instrument for everything else.
- Severity. Plate counts correlate loosely with severity and are not a grade. An anomaloscope grades; plates do not.
- Dichromacy versus anomaly. Whether a cone is absent or merely mistuned is largely beyond it.
- Blue-yellow deficiency. The classic series barely tests the tritan axis at all, it was built for red-green screening. Tritan screening usually needs a different test, such as the Farnsworth D-15.
- Acquired deficiency. Plates assume a stable congenital pattern and are insensitive to the early, patchy losses typical of disease.
This is why an optometrist who sees a failed plate test does not stop there. Plates point; an anomaloscope, a D-15 arrangement test or an occupational lantern test decides.
What This Site Does Differently
The 1917 plates are still in copyright, Kanehara Shuppan publishes them, and reproductions circulating online are usually scanned, recompressed and color-shifted enough that the confusion lines no longer land where Ishihara put them. A scan that has been through a JPEG encoder and an unknown color profile is not the test any more; it just looks like it.
So the plates here are generated instead. For each plate the figure and background colors are computed in LMS cone space, the same space as the simulator, by separating them along the exact cone axis the target deficiency is missing. That makes the two collapse onto an identical color under simulation by construction, rather than by having been printed that way in 1917 and hopefully surviving the trip.
Three things follow from that:
- Every plate is checked automatically before it is shown. A plate whose figure does not actually vanish under simulation is rejected and regenerated.
- Every visitor gets a different dot arrangement, so the answers cannot be memorised or looked up.
- Nothing copyrighted is reproduced.
What generation cannot fix is your screen. The plates are correct; the display showing them is uncalibrated, the room is unmeasured, and neither was true of a printed plate under a daylight lamp. That gap is the reason every result here is a screening and not a diagnosis.
Take It
The color blind test on this site runs fourteen plates in about ninety seconds: twelve vanishing plates covering the protan, deutan and tritan axes four times each, and two demonstration plates that check your screen is not distorting the colors. If the demonstration plates come out wrong, the run is not scored, because at that point the screen is the thing being measured.
Then read what the types mean, or go straight to protanopia, deuteranopia or tritanopia.