Tritanopia is the rare one. Almost every explainer opens by saying so, then files it under blue-yellow color blindness, offers a picture of a muddy sunset, and moves on to the red-green types that affect far more people. That summary is short enough to be memorable and wrong enough to be worth correcting, because the thing tritanopia actually removes is not blue.
So I measured it. I built the standard dichromat simulation, ran the whole sRGB cube through it three times, once for each missing cone type, and compared what came out. The headline result is that blue is one of the best preserved colors on the screen for someone with tritanopia, and yellow is the one that stops existing. Pure yellow comes out of the simulation as #FFF0F3, which is a white with a faint pink cast, 36.2 CIEDE2000 away from where it started.
How the simulation works, and why you can trust the numbers
You cannot ask a tritanope to describe a color and get a usable answer, because they have nothing to compare it to. The standard workaround is a geometric one, set out by Brettel, Viénot and Mollon in 1997. Convert each color into the responses of the three cone types, then note that a dichromat has only two of them. Every color that differs only in the missing cone's response collapses onto the same percept, so the whole three-dimensional color solid folds down onto a surface. The surface is anchored on two constraints: greys must stay grey, because a dichromat still agrees with everyone else about white, and two reference wavelengths that both eyes are known to see the same way must stay put.
For tritanopia those anchors are 485 nm and 660 nm. For the red-green types they are 475 nm and 575 nm. The surface is two half-planes hinged along the neutral axis rather than one flat plane, which is the detail the cheaper simulations skip and the reason their yellows come out wrong.
My implementation uses the Hunt-Pointer-Estevez cone fundamentals normalised to D65, so the simulation is neutral-preserving on an ordinary sRGB display. The obvious sanity check is the grey ramp: every neutral from black to white should pass through untouched. Across the full ramp the largest error was 0.000 CIEDE2000. Every color difference quoted below is CIEDE2000, the formula I use for everything on this site and explain at length in what CIEDE2000 measures. The threshold I treat as the edge of noticing is 2.3, which comes from the just noticeable difference work.
The name is backwards: yellow moves twice as far as blue
I walked a fully saturated hue ring, one color every 15 degrees, and measured how far tritanopia pushes each one. Averaged across the blue band from 195 to 255 degrees, the shift is 17.1 CIEDE2000. Averaged across the yellow band from 30 to 75 degrees, it is 34.6. Yellow moves almost exactly twice as far.
Cyan barely moves at all. Pure #00FFFF simulates to #43F9FF, a shift of 2.9 CIEDE2000, which is a hair above the threshold where anyone notices anything. A tritanope and a trichromat looking at the same cyan are, for practical purposes, looking at the same color. Deep sky blue shifts 6.6. Pure blue shifts 20.8, which is real but modest next to what happens at the other end.
#FFFF0036.2
#FFD70040.5
#FFA50036.8
#FF00FF53.6
#80008042.5
#00FFFF2.9
#00BFFF6.6
#0000FF20.8
#FF000014.6
#00800025.4
The biggest single shift in the whole ring is not in the yellows at all. It is at 285 degrees, the violet between purple and magenta, which moves 54.1 CIEDE2000 and lands on a dull orange brown. Magenta itself goes to #EB7200. That is a larger displacement than anything either red-green type produces anywhere on the ring, and it is the part of tritanopia that nobody puts in the summary. Violet turning orange is a stranger failure than yellow turning white, and it happens because violet is the one hue on the screen that leans hardest on the short-wave cone to exist at all.
Six near-whites that become yellow
Abstract shifts are easy to nod at, so I ran the 139 distinct CSS named colors through all three simulations and counted the pairs that go from clearly different to a normal observer, more than 5 CIEDE2000 apart, to indistinguishable afterwards, under 2.3.
Tritanopia produced 59 newly confusable pairs. Protanopia produced 52 and deuteranopia 50. The supposedly mild deficiency creates the most collisions in the standard web palette, which was not the result I expected.
What is more interesting is which pairs. Six of the pale, warm near-whites in the CSS list collapse onto pure yellow:
- seashell and yellow, 29.6 apart normally, 2.00 under tritanopia
- linen and yellow, 28.5 becomes 1.33
- old lace and yellow, 26.8 becomes 1.67
- papaya whip and yellow, 24.7 becomes 1.74
- cornsilk and yellow, 23.6 becomes 2.19
- lemon chiffon and yellow, 20.0 becomes 2.25
That is the practical shape of the deficiency. It is not that yellow looks like a different color, it is that yellow stops being a color and joins the family of off-whites. A yellow highlight on a white page, a yellow warning band, a yellow line on a chart against a cream background: all of these are doing no work at all.
The other tritan collisions worth naming are royal blue against teal, 31.7 apart normally and 0.98 afterwards, which is the sharpest single collapse in the set; grey against medium purple at 26.1 to 1.76; and dark orange against salmon at 22.8 to 1.32. Compare that with what protanopia does to the same list, where the widest collision is plain green against plain red, 72.2 CIEDE2000 apart to a trichromat and 1.17 apart to a protanope. Red-green blindness produces fewer collisions but much more dramatic ones.
Where the confusion actually sits on the hue circle
Lightness passes through every dichromacy untouched, so any measurement that mixes light colors with dark ones flatters the simulation. To isolate the chromatic loss I drew 400,000 pairs of colors matched to within 1 unit of L*, kept only the pairs a normal observer would call obviously different at 10 CIEDE2000 or more, and asked how many fall below the 2.3 threshold after simulation.
Across all hues, tritanopia collapses 2.56 percent of those pairs, protanopia 2.44 percent and deuteranopia 1.03 percent. The rare deficiency confuses the most, once you take lightness out of the picture. Broken down by the hue of the pair, the pattern is sharper still.
Tritanopia has the single deepest spike of the three, 9.3 percent in the 150 to 180 degree bin, which in CIELAB terms is the green-cyan wedge that holds teal, sea green and turquoise. Its second peak, 6.4 percent, sits at 240 to 270 degrees in the blue-violets. Protanopia spikes almost as hard, 9.2 percent, but 60 degrees away in the yellow-greens. Deuteranopia never exceeds 2.8 percent in any bin: it spreads its losses thinly across half the circle instead of dropping them all in one place.
That difference in shape matters more than the totals. A deficiency that loses a little everywhere is one you compensate for with general caution. A deficiency that loses almost nothing except in one narrow wedge is one that ambushes you, because ninety percent of the time your color judgement is fine and then a teal and a royal blue turn out to be the same thing. The wedge tritanopia takes is exactly the region I wrote about in the piece on shades of teal, where teal, turquoise and aqua already sit closer together than their separate names suggest.
The contradiction: fewer colors lost, more pairs confused
Here is the result that took the longest to make sense of. I counted how many mutually distinguishable colors survive each deficiency, by greedily packing a 16 by 16 by 16 sample of sRGB so that no two kept colors are closer than 2.3 CIEDE2000. A normal observer keeps 2,506 of them. Then the same count on the simulated versions:
- protanopia: 414, or 16.5 percent of normal
- deuteranopia: 413, or 16.5 percent
- tritanopia: 748, or 29.8 percent
Tritanopia keeps 1.81 times as many distinguishable colors as either red-green type. So in raw inventory it really is the milder deficiency, which is what the reputation says. But it also confuses more isoluminant pairs than either of them, which is the opposite of what the reputation says. Both numbers came out of the same simulation, so one of my intuitions had to be wrong.
The explanation is not about how much color is left but about what the remaining color is correlated with. When you lose a cone, one chromatic axis survives. For protanopia and deuteranopia the survivor is the blue-yellow axis. For tritanopia it is red-green. My first guess was that the red-green axis is simply longer in sRGB, but measuring the gamut says otherwise: the surviving axis spans 198 units for protanopia and only 137 for tritanopia.
The real difference is redundancy with lightness. I projected each simulated gamut onto its own surviving chromatic direction and correlated that against L*:
- protanopia: r = -0.505, so 25.5 percent of the surviving chromatic axis is already predictable from lightness
- deuteranopia: r = -0.393, 15.5 percent predictable
- tritanopia: r = -0.244, only 5.9 percent predictable
Blue is dark and yellow is bright, in sRGB and in life. A quarter of what the blue-yellow axis tells a protanope, lightness had already told them. The red-green axis carries almost no lightness information at all, which is why red and green of matched brightness are the classic confusion pair for everyone else and why a tritanope, who keeps that axis intact, ends up with more genuinely independent color directions from a shorter axis. Shorter but non-redundant beats longer but duplicated. This is the same opponent structure I went through in opponent process theory, seen from the side where one channel is missing.
One more number in the same direction. Between isoluminant pairs, mean color difference for a normal observer across my sample is 31.61 CIEDE2000. Tritanopia retains 76.5 percent of that, protanopia 71.1 and deuteranopia 71.2. On average the tritanope sees more difference. It is only in the specific wedge where the short-wave cone was doing all the work that they see almost none.
Tritanopia does pay for it in saturation. Across the fully saturated hue ring it retains 51.9 percent of the original chroma, against 62.8 percent for deuteranopia and 67.6 for protanopia. The tritanope's world has more distinct colors in it and every one of them is duller.
Why it is rare, and why acquired cases are not
Congenital tritanopia comes from variants in OPN1SW, the gene for the short-wave cone opsin, which sits on chromosome 7 rather than the X chromosome. That single fact explains the whole epidemiology. Red-green deficiency is X-linked recessive, so it lands on roughly 8 percent of men and 0.5 percent of women. Tritanopia is autosomal dominant with incomplete penetrance, which means it affects men and women equally and turns up in around 1 in 10,000 people, and that some carriers of the variant never show the phenotype at all. For a long time researchers doubted congenital tritanopia existed, because so few cases had ever been written up.
Acquired tritan defects are a different story and much more common. The short-wave cones are the sparsest in the retina, they are absent from the very centre of the fovea, and the lens yellows with age, which filters exactly the wavelengths those cones depend on. Clinicians have a rule of thumb for this, usually credited to Köllner: disease in the outer retina tends to produce blue-yellow defects, disease in the optic nerve tends to produce red-green ones. Glaucoma, diabetic retinopathy and ordinary ageing all push vision in the tritan direction. If you have ever noticed that an older relative is oddly confident that a navy garment is black, you have probably watched a mild acquired tritan defect at work.
This is worth separating from the congenital case, because the numbers above describe a complete absence of the short-wave cone. A mild tritanomaly, where the cone is shifted rather than missing, produces a weaker version of the same pattern: the same wedge, less deep.
What this means if you design or test with color
The practical advice usually given for color blindness is aimed at the red-green types and does not transfer. Three things fall out of the measurements above.
Yellow on white is not a signal
Six named near-whites collapsing onto pure yellow is not an edge case, it is a description of how highlighters, warning bands and chart gridlines usually get coloured. Yellow carries almost all its identity in the short-wave channel. If it must mean something, give it a border, a shape or a lightness step as well.
Teal and blue are the pair to check
Royal blue against teal was the sharpest collapse I found, from 31.7 to 0.98 CIEDE2000. The 150 to 180 degree wedge is where a tritanope loses almost 1 pair in 10 at matched lightness. If a palette distinguishes two states using teal and blue, it distinguishes nothing.
Purple is not a safe accent
Violet at 285 degrees produced the single largest displacement in the entire study, 54.1 CIEDE2000, and it lands in the oranges. Grey and medium purple become confusable at 1.76. Purple reads as sophisticated and neutral to a trichromat, and to a tritanope it is neither.
If you want to see any of this rather than read it, the color blindness test on this site walks through the plate patterns for all the deficiency types, including the tritan plates that most free tests leave out. And the broader question of what a two-cone world looks like from the inside is one I came at from a different angle in what colors dogs can see: a dog is a dichromat too, but the opposite kind, keeping the blue-yellow axis and losing red-green. A dog and a tritanope are looking at complementary halves of the same missing information.
The short version
Tritanopia is called blue-yellow color blindness and blue is the part of it that survives. Cyan shifts 2.9 CIEDE2000, which nobody would notice. Yellow shifts 36.2 and lands among the off-whites, violet shifts 54.1 and lands in the oranges. The deficiency keeps 1.81 times as many distinguishable colors as red-green blindness, because the axis it holds onto carries information lightness does not already supply, and it still confuses more pairs at matched lightness than either red-green type, because the losses are concentrated in one narrow wedge instead of spread thin.
Whatever your cone complement, the gap between telling two colors apart when they sit side by side and pinning one down from memory is a separate problem, and a larger one. That is the gap the Color Memory Game is built on, and it does not close for anyone.