The color wheel: measured, and it is not a circle

The color wheel is probably the most reproduced diagram in the whole of art education. Twelve wedges, red at the top, primaries at the thirds, every neighbour a tidy 30 degrees away, and every color facing its opposite across the middle. It is printed on plastic discs sold in every art shop, it is the first thing taught in a design foundation course, and it is the mental model almost everyone uses when they pick colors.

I wanted to know whether it survives being measured. So I built the wheel from a published paint-mixing model rather than eyeballing hex codes off a poster, converted all twelve spokes into CIELAB, and checked the three things the diagram claims: that the steps are equal, that it is a circle, and that colors sit opposite each other.

None of the three held. The gaps between neighbouring spokes range from 2.16 degrees to 84.90, a factor of 39. The rim swings from a chroma of 21.54 to 104.55, so the shape is a lopsided blob rather than a circle. And the average pair the wheel calls opposite is 54 degrees away from actually being opposite, with red and green, the most famous pair of the lot, sitting about 94 degrees apart.

The part I did not expect is what happened when I ran the same three tests on the color wheel built into design software. It came out worse.

How I built the wheel

The hard part of measuring a color wheel is getting an honest one to measure. Search for a color wheel image and you will find a thousand versions that disagree with each other, because most of them are drawn by hand and the tertiary colors in particular are wherever the illustrator felt like putting them.

So I generated the wheel instead. Gossett and Chen published a red-yellow-blue color space for visualization work, defined by the eight corners of an RYB cube given in RGB, with trilinear interpolation between them. It is a proper subtractive-feeling model: mix their red and yellow and you get orange, mix their yellow and blue and you get a green rather than the grey that additive mixing would hand you.

To get twelve spokes I walked the perimeter of the RYB triangle in twelve equal steps, red to yellow to blue and back to red, and converted each one through the cube. That gives the traditional wheel with no illustrator in the loop: three primaries, three secondaries at the midpoints, six tertiaries between them. Every color below came out of that walk.

Measurement is CIELAB under D65, hue angle and chroma in the usual polar form, and perceptual distances in CIEDE2000, the same method the rest of the pieces on this site use. Because CIELAB is known to have hue-linearity problems of its own, particularly in the blues, I re-ran every conclusion in Oklab as a check. The numbers shrink. None of them go away. I have given both wherever it matters.

The twelve spokes

Here is the wheel, generated and measured.

Read down the hue column and the problem is already visible before any arithmetic. Red is at 40 degrees. Red-orange, one full step of the wheel later, is at 42.16. Meanwhile blue-green at 189.16 is followed by blue at 274.06, and that step is worth 85 degrees.

The steps are not 30 degrees

Each bar below is the measured hue gap between one spoke and the next. The thin vertical line is 30 degrees, which is where every bar would end if the diagram were telling the truth.

• Smallest step: 2.16 degrees, red to red-orange.
• Largest step: 84.90 degrees, blue-green to blue.
• Ratio between them: 39 to 1. Standard deviation around the promised 30 degrees: 21.69.
• In Oklab the same spread comes out 5.24 to 64.11, a ratio of 12 to 1. Less extreme, still nothing like even.

Group the spokes into the three thirds the primaries are supposed to define and the imbalance gets easier to hold in your head. Each third should be 120 degrees of hue. Red round to yellow is 62.85 degrees, just over half its share. Yellow round to blue is 171.20, nearly a degree and a half for every degree it is owed. Blue back to red is 125.94 and very nearly honest.

Red and red-orange are the same color

Two degrees of hue is not a distinction. It is inside the noise of anybody’s vision, and well inside the printing tolerance of the poster the wheel is drawn on. Those two spokes differ mostly in lightness and chroma: red is L* 53.24 at chroma 104.55, red-orange is L* 60.25 at chroma 83.26. The wheel presents them as a step around the hue circle when what has actually happened is that one of them got paler.

This is not a quirk of my particular model. It is the reason that the warm quadrant of every color wheel looks crowded: sRGB red sits at a CIELAB hue of 40 degrees and full-strength orange is at 65, so the entire visual distance from red through to a proper orange is 25 degrees of hue, and the wheel spends three of its twelve spokes crossing it.

There is a hole where cyan should be

The 84.90 degree jump from blue-green to blue is 23.6 percent of the circle. Almost a quarter of all the hues there are fall into a gap between two adjacent spokes, and no spoke of the artist’s wheel represents any of them.

Here is what is living in the gap.

That strip is cyan, then teal running into azure, then the vivid sky-blues. It contains most of what people mean when they say a color is turquoise. It contains the color of a swimming pool, of shallow tropical water, of the sky about halfway up. The traditional wheel represents the whole of it with one muted spoke labelled blue-green at a chroma of 21.54, the weakest color on the entire rim.

I think this is the single most consequential thing wrong with the diagram, and it has a straightforward cause. The wheel is built from paint primaries, and its blue primary is not a cyan. Gossett and Chen’s blue corner is a deep ultramarine. Mix a warm ultramarine with a yellow and the result runs to a leafy green without ever passing through cyan, because there is no cyan in either ingredient. Printing worked this out a century ago and switched its blue primary to cyan, which is exactly why CMYK can reach colors that a box of paints cannot. The wheel kept the old primary and inherited the hole.

It is not a circle

The second claim the diagram makes is geometric. Drawing all twelve colors at the same distance from the centre says they are equally colorful, which is what lets you treat the wheel as a compass where only the angle matters.

Plot each spoke at its measured hue angle and its measured chroma and the compass falls apart.

• Chroma runs from 21.54 at blue-green to 104.55 at red. The rim is nearly five times further out on one side than the other.
• Lightness runs from 39.56 at blue to 97.14 at yellow, a spread of 57.58. Yellow is not a little brighter than blue, it is more than twice as light.
• Six of the twelve spokes are bunched into the 103 degrees between red and yellow-green. The other six are spread across the remaining 257.

The pinch at blue-green is worth looking at directly, because it is the same fact as the hole, seen from a different angle. That spoke is not just badly placed, it is barely a color at all. At chroma 21.54 it is closer to a grey-green than to anything a swatch book would call turquoise. Everything the model had left after mixing a warm blue into a yellow was a muddy middle, and the wheel dutifully gives it a twelfth of the circle.

If the uneven brightness feels familiar, it is the same effect that makes yellow text unreadable on white and blue text hard to read on black, and it is a large part of why warm and cool is such a stubborn intuition. Warm colors on this wheel really are lighter, by an average of 20 points of L*.

Nothing on the wheel is opposite anything

The third claim is the one people actually use. Pick a color, go straight across, and the diagram hands you its complement. Six spoke pairs, each supposedly 180 degrees from its partner.

Not one pair lands on 180. The average miss is 54.14 degrees and the worst is 86.12. In Oklab the average miss is 43.13, so the effect is smaller but the verdict is identical: the wheel is not measuring opposition, it is measuring position on a diagram that was drawn to look tidy.

Red and green are the case worth dwelling on, because that pair does more work in design and in Christmas decoration than any other. They come out 93.88 degrees apart in CIELAB and 107.71 in Oklab. That is a quarter turn, not a half turn. Red’s actual opposite in hue terms lands at 220 degrees, in the middle of the arc the wheel does not represent, which is to say that red’s complement is one of the missing cyans. The hole and the broken opposites are the same defect.

There is a separate and stronger sense in which red and cyan are opposites, which is that they cancel each other in the visual system rather than on a mixing diagram. That is opponent process territory, and it is why the afterimage of a red square is cyan and not green. Two independent lines of evidence, the geometry and the physiology, put the complement of red in the same place, and the artist’s wheel puts it somewhere else. Our longer piece on complementary colors works through what that means for a palette in practice.

The wheel in your software is worse

At this point the obvious response is that the paper wheel is a pre-scientific artefact and the fix is to use the color wheel in a design tool, which is built from actual RGB numbers.

So I ran the same three tests on it: twelve fully saturated HSV hues at 30 degree intervals, the wheel that sits in the color picker of almost every piece of software you own.

• Hue gaps run from 5.39 to 88.64 degrees. Better ratio than the paper wheel, but the standard deviation around 30 is 21.96, marginally worse.
• Lightness spread is 64.84 against the paper wheel’s 57.58. Worse.
• The real gap is in how far you travel per step. On the paper wheel a 30 degree turn moves you between 7.84 and 29.70 CIEDE2000 units, a ratio of 3.79. On the HSV wheel it is 5.19 to 44.05, a ratio of 8.49. Two and a quarter times less even.

That last number is the one I keep turning over. The HSV wheel has four adjacent pairs that are separated by less than a quarter of what separates cyan from azure, and one step, #00FFFF to #0080FF, that crosses 44 units of perceptual distance on its own. The paint wheel, assembled by people with no access to colorimetry who were working entirely from what looked evenly spaced, distributes perceived difference around the circle noticeably better than the one derived from arithmetic on the RGB cube.

That is not a fluke of good taste. HSV was designed in the 1970s to be cheap to compute, and even spacing was never one of its goals. It inherits whatever the phosphors of a display happen to do. The artist wheel was iterated on for two centuries by people whose only test was whether it looked right. On the one criterion that is about looking, the visual iteration wins.

What an even wheel looks like

For contrast, here is a wheel with twelve spokes at genuinely equal perceptual intervals. It is built the boring way: fix lightness at L* 60, fix chroma at 38, and step the CIELAB hue angle by exactly 30 degrees.

A ratio of 1.63 against the paper wheel’s 3.79 and HSV’s 8.49. The steps really are close to equal, the missing cyans are back in, and two of the spokes have to be clipped because sRGB cannot hold that much cyan at that lightness in the first place.

And it looks dreadful. Every color is dusty, because holding chroma constant means dragging every hue down to what the weakest hue can manage. There is no red, only a dull rose, because sRGB red at chroma 38 is a rose. You could not teach with this and no art shop would sell it.

Which is the honest end of the argument. Perceptual uniformity and looking like the colors you actually want are in direct conflict, and the traditional wheel picked the second one. Pinning it to a Munsell-style grid, where each hue is allowed its own maximum chroma, is the compromise that works, and it is also why Munsell’s solid is a lumpy tree rather than a cylinder.

So is the color wheel wrong

It is wrong about the three things I tested, and it was never really claiming them. Every number above measures the wheel as a map of human color perception. That is not what it is. It is a mixing diagram, and the question it answers is what comes out of the tray when you combine two things that are already in it.

Judged on that, it does well. Red next to yellow really does give you the orange the wheel puts between them. Sitting opposite on the wheel really is a decent predictor that two paints will dull each other toward grey, which is the practical thing a painter wants from a complement, and which is a pigment-mixing fact rather than a perceptual one. The wheel is a compressed record of what happened when people mixed paint, and it is accurate about paint.

The trouble starts when it gets used as a perceptual map, which is exactly what happens whenever it is applied to screens. Take a color scheme off the wheel and build a website with it and you have imported a model of pigment behaviour into a medium where light is added rather than subtracted, which is a different physical process with a different geometry. The palette that comes out is usually defensible, because the wheel encodes a lot of accumulated taste, but it is defensible for reasons that have nothing to do with the derivation.

My own view, after doing the measurement, is that the wheel deserves more respect as a piece of empirical work than it usually gets, and much less authority as a theory than it usually claims. Keep it for mixing. Reach for a scheme built from measured hue when you are choosing colors for a screen. And treat the phrase “opposite on the color wheel” as a description of a diagram, never as a fact about color.

If you want to feel the unevenness

Reading these numbers is one thing. The uneven spacing is much easier to believe once you have failed at it yourself, and there are two drills on the site that surface it fast.

• Hue Sort gives you a scrambled set of colors to put back in order around the circle. The arcs where the wheel is crowded are the arcs where everybody slows down, and you can watch yourself confidently misplace reds against oranges by the same couple of degrees the measurement above found.
• Mixer asks you to hit a target by combining components, which is the wheel’s actual job. It is the fastest way to learn that mixing toward a hue and moving toward it on a diagram are different motions.

The main color memory game is the broader version of the same exercise: hold a color in your head, then reproduce it. If you want the structured route through all of it, our guide to training your eye for color sets out the drills in order, and the primary colors piece covers why the three at the corners of this wheel are not the only candidates for the job.

Method notes

Reproducing this needs nothing exotic. The RYB cube corners are in Gossett and Chen’s paper; trilinear interpolation over them is eight weighted lookups. The twelve spokes are an equal-step walk around the R to Y to B perimeter. sRGB to CIELAB is the standard D65 pipeline with the sRGB transfer function, and CIEDE2000 follows the CIE formulation.

Three caveats I want on the record. The tertiary colors depend on the mixing model, so a different RYB formulation would shift them by a few degrees, though not enough to close an 85 degree gap. CIELAB is not perfectly hue-linear, worst of all in the blues, which is precisely where the largest gap sits, and that is why every headline claim is also given in Oklab. And the whole thing lives in sRGB, so the chroma ceilings are the ceilings of a display rather than of pigment or of the eye.

What survives all three caveats is the shape of the answer: the steps are unequal by a large factor, an arc containing cyan is missing, and the pairs drawn opposite each other are not opposite in any measurable sense.