Color theory is the set of rules for which colors go together and why. It has three moving parts: a wheel that arranges hues in a circle, a vocabulary for describing a color once you have found it, and a handful of named schemes that pick colors off the wheel at fixed angles. Learn those and you can build a palette without guessing.
The trouble is that the wheel is a geometric object and your eye is not. The wheel says the gap from green to blue-green is the same size as the gap from orange to yellow, because both are 30 degrees. Your eye disagrees, and it disagrees by a lot.
This site scores color guesses with CIEDE2000, the standard formula for how different two colors look to a human observer. That gives a convenient way to audit color theory rather than repeat it. I ran the whole wheel through the formula one degree at a time and measured every scheme across all 360 rotations. The short version: the twelve stations of the standard wheel are separated by anywhere from ΔE 5.2 to ΔE 43.9, an eight-fold spread, and almost all of the compression sits in one place. Below is what color theory teaches, which parts hold up under measurement, and a rebuilt twelve-station wheel whose stations are genuinely even.
What color theory actually consists of
Strip away the terminology and there are only three ideas.
1. The color wheel
Hue is circular. Red shades into orange, orange into yellow, yellow into green, and if you keep going you come back to red through the magentas. There is no natural start or end, so a circle is the honest shape for it. Isaac Newton drew the first one in 1704 after splitting sunlight with a prism and noticing that the two ends of the spectrum, red and violet, looked like neighbours despite sitting at opposite ends of the physics.
There are two wheels in common use and they are not the same wheel. The RYB wheel is the artist's one, built on red, yellow and blue as primaries, and it is what most color theory courses teach. The RGB wheel, and the HSV hue circle derived from it, is what screens and design software use, built on red, green and blue. On the artist wheel, red sits opposite green. On the screen wheel, red sits opposite cyan. Both claim to show complements. They cannot both be right, and the section on complements below puts numbers on which one is closer.
2. The three properties of a color
Any color you can name breaks into hue, saturation and value. Hue is the position on the wheel, measured in degrees. Saturation is how much of the hue is present rather than gray. Value is how light or dark it is. Color theory spends nearly all of its attention on hue and treats the other two as adjustments, which turns out to be the wrong weighting. More on that shortly.
If you want the notation, the six digits of a hex code encode the same color in a different coordinate system, and how to read hex color codes walks through the conversion.
3. The named schemes
A scheme is a rule for picking hues off the wheel at set angles. The whole family, by the size of the angle:
- Monochromatic, 0 degrees. One hue, varied by saturation and value.
- Analogous, a cluster inside 60 degrees. Neighbours on the wheel.
- Triadic, 120 and 240 degrees. Three hues at even thirds.
- Split complementary, 150 and 210 degrees. A base plus the two neighbours of its opposite.
- Complementary, 180 degrees. A base and its opposite.
- Tetradic, 0, 90, 180 and 270 degrees. Two complementary pairs at right angles.
There is also the vocabulary layer: primary colors are the ones you cannot mix, secondary are what you get from mixing two primaries, and tertiary colors are the six in-between hues that fill the wheel out to twelve stations.
Where the wheel stops matching your eye
Here is the measurement that reframes everything else. Take the twelve stations of the standard wheel at full saturation and brightness, and measure the perceptual distance from each one to the next:
- Red to orange, hue 0 to 30: ΔE 21.2
- Orange to yellow, 30 to 60: ΔE 42.4
- Yellow to chartreuse, 60 to 90: ΔE 18.5
- Chartreuse to green, 90 to 120: ΔE 5.2
- Green to spring green, 120 to 150: ΔE 9.0
- Spring green to cyan, 150 to 180: ΔE 27.3
- Cyan to azure, 180 to 210: ΔE 43.9
- Azure to blue, 210 to 240: ΔE 28.7
- Blue to violet, 240 to 270: ΔE 9.2
- Violet to magenta, 270 to 300: ΔE 23.0
- Magenta to rose, 300 to 330: ΔE 20.0
- Rose to red, 330 to 0: ΔE 26.2
The mean is ΔE 22.9. The largest step is 8.4 times the smallest. Every one of those gaps is exactly 30 degrees of hue, and they are not remotely the same size to look at. A step of ΔE 5 is at the edge of what an untrained observer notices side by side. A step of ΔE 44 is two obviously unrelated colors.
Zoom out and the pattern is cleaner. Walking the whole wheel one degree at a time and adding up the distances gives a total path length of ΔE 277.9. Split into quarters:
- Red to chartreuse, 0 to 90 degrees: ΔE 81.3, or 29 percent
- Chartreuse to cyan, 90 to 180 degrees: ΔE 41.8, or 15 percent
- Cyan to blue, 180 to 270 degrees: ΔE 85.1, or 31 percent
- Blue to red, 270 to 360 degrees: ΔE 69.7, or 25 percent
A quarter of the wheel by geometry is 15 percent of it by perception. The entire green region, from chartreuse through green and spring green to the edge of cyan, is perceptually about half the size the wheel draws it. Meanwhile the cyan-to-blue quarter is doing twice the work per degree.
The extreme case is worth stating on its own. A 30 degree step starting at hue 100, in the greens, is worth ΔE 3. The same 30 degree step starting at hue 181, just past cyan, is worth ΔE 44.2. Identical rotation, 14.8 times the perceptual result. When a guide tells you to move a hue "a little" or "a lot" in degrees, it is using a unit that changes value depending on where you are standing.
None of this is a flaw in color theory so much as a flaw in reading it literally. The wheel is a mixing diagram, and it is a good one. It stopped being a perception diagram the moment anyone assumed equal angles meant equal differences. Why the compression sits where it does comes down to how the eye weights the three cone types, which what is CIEDE2000 covers in detail, and how many colors can you see covers the resolution limits behind it.
A perceptually even twelve-station wheel
If ΔE 277.9 is the full circumference, then twelve genuinely even stations sit ΔE 23.2 apart. Solving for those positions gives a wheel that looks like the familiar one in places and drifts badly in others:
- Station 1: hue 0, #ff0000. Nominal 0, on the nose.
- Station 2: hue 32, #ff8800. Nominal 30.
- Station 3: hue 48, #ffcc00. Nominal 60, drifted 12 degrees.
- Station 4: hue 69, #d9ff00. Nominal 90, drifted 21 degrees.
- Station 5: hue 145, #00ff6a. Nominal 120, drifted 25 degrees the other way.
- Station 6: hue 175, #00ffea. Nominal 150, drifted 25.
- Station 7: hue 190, #00d5ff. Nominal 180.
- Station 8: hue 205, #0095ff. Nominal 210.
- Station 9: hue 220, #0055ff. Nominal 240, drifted 20.
- Station 10: hue 271, #8400ff. Nominal 270.
- Station 11: hue 302, #ff00f7. Nominal 300.
- Station 12: hue 334, #ff006f. Nominal 330.
Read that list from station 4 to station 9 and you can see the whole problem in one place. Between hue 69 and hue 220, a span of 151 degrees on the standard wheel, the even wheel fits only five stations. The greens and cyans collapse. Everywhere else the drift is under five degrees, which is a useful thing to know: the standard wheel is close to honest across the reds, oranges, violets and magentas, and it is only in the green-through-blue arc that it misleads.
There is one residual unevenness in that list. The arc from station 9 to station 10, hue 220 to 271, measures ΔE 14.5 against a target of 23.2, because the blues are so compressed that no single degree solves it cleanly. Even a wheel built specifically to be even cannot fully even out the blue band.
Hue is not the biggest lever
Color theory is overwhelmingly a theory of hue. Schemes are hue angles, the wheel is a hue circle, the vocabulary is mostly hue names. So it is worth asking how much of a color's appearance hue is actually responsible for.
I took a grid of colors across the wheel at three saturations and three values, and moved each one by a comparable amount along each of the three axes: 30 degrees of hue, 30 points of saturation, 30 points of value. Mean perceptual cost of each move:
- 30 point drop in value: ΔE 19.9
- 30 degree rotation of hue: ΔE 16.5
- 30 point drop in saturation: ΔE 10.3
Value moves the color furthest, hue is second, saturation is a distant third at roughly half the effect of value. That ordering is close to invisible in most color theory writing, which will spend three chapters on hue relationships and one paragraph on tints and shades.
For anyone playing a color matching game this has an immediate consequence. If a guess is wrong and you have limited moves to fix it, checking lightness first is a better use of them than debating the hue, because lightness is where the largest correctable error hides. Training your eye for color works through the drills for separating the three axes, and color memory versus color matching covers why the two skills come apart.
Two wheels, two sets of complements, one winner
Back to the disagreement between the artist wheel and the screen wheel. On the RYB wheel, red pairs with green, orange with blue, yellow with violet. On the HSV wheel, a complement is whatever sits 180 degrees away, full stop. Measuring both:
- Red and green, hues 0 and 120: ΔE 86.6
- Orange and blue, hues 30 and 240: ΔE 68.1
- Yellow and violet, hues 60 and 285: ΔE 95.9
- Yellow-green and red-violet, hues 90 and 315: ΔE 102.2
- Blue-green and red-orange, hues 180 and 15: ΔE 67.0
- Blue-violet and yellow-orange, hues 270 and 45: ΔE 80.7
The six artist pairs average ΔE 83.4 and run from 67 to 102, a spread of 35. Every 180 degree pair on the HSV wheel averages ΔE 91.9 and runs from 53 to 112, a spread of 59.
So the screen wheel wins on raw contrast by about 10 percent, and the artist wheel wins on consistency by a wide margin. Pick a complement at random on the artist wheel and you are guaranteed something between ΔE 67 and ΔE 102. Do the same on the screen wheel and you might land on ΔE 112, or you might land on hue 24 against hue 204, orange #ff6600 against azure #0099ff, which measures ΔE 53 and is the weakest complementary pair anywhere on the circle. It is a perfectly pleasant pairing. It is not the maximum-contrast relationship the theory promised.
The strongest is hue 103 against hue 283, chartreuse #48ff00 against purple #b700ff, at ΔE 112. Worth noting that the strongest complement on the wheel is built from a hue sitting in the most compressed region of it. Compression along the wheel and distance across it are separate things.
The practical read: the artist wheel is not a crude approximation of the screen wheel that we keep teaching out of habit. Its pairings were chosen by people looking at pigments, and they are measurably more reliable than the geometric alternative, if slightly less extreme. That is a defensible trade.
What each scheme actually buys you
Every scheme measured across all 360 rotations at full saturation and brightness. Mean is the average distance between the colors in the palette, floor is the average weakest pairing, and worst floor is the weakest pairing found anywhere on the wheel:
- Complementary, 2 colors, 1 pair. Mean ΔE 91.9. Floor 91.9. Worst floor 53.
- Triadic, 3 colors, 3 pairs. Mean ΔE 69.1. Floor 50.0. Worst floor 39.4.
- Split complementary, 3 colors, 3 pairs. Mean ΔE 67.4. Floor 38.5. Worst floor 11.6.
- Tetradic square, 4 colors, 6 pairs. Mean ΔE 68.1. Floor 34.6. Worst floor 24.9.
- Rectangle at offset 60, 4 colors, 6 pairs. Mean ΔE 67.3. Floor 30.5. Worst floor 11.6.
- Analogous at 0, 30 and 60, 3 colors, 3 pairs. Mean ΔE 28.4. Floor 15.6. Worst floor 3.0.
The interesting column is the floor, not the mean. Once you go past two colors, every scheme lands in a narrow band around ΔE 67 to 69 for its average. Triadic, split complementary, tetradic and rectangular palettes are effectively identical on that measure. What separates them is how weak their weakest link gets. Triadic holds a floor of 50 and never drops below 39 anywhere on the wheel, which is why it is the most forgiving three-color scheme to build blind. Split complementary has a nearly identical average and a floor 12 points lower, with a worst case of 11.6 where the two accents nearly merge.
And adding a fourth color buys nothing in spread. The square tetrad averages ΔE 68.1 across six pairs, which is fractionally below what a triad manages across three, while its floor drops from 50 to 34.6. More colors, same average separation, weaker weak point. That is the case for restraint stated as a number rather than as taste.
What a scheme is worth as a guess
One more angle, specific to playing rather than designing. If a target color comes up and you answer with a color that sits at a scheme relationship to it, what does that score out of 10 on this site's scale, averaged over the whole wheel?
- 30 degrees off, an analogous neighbour: 5.90. Best case 9.65, worst 2.96.
- 60 degrees off, the far edge of analogous: 3.58.
- 120 degrees off, a triadic partner: 1.71.
- 150 degrees off, a split complementary accent: 1.34.
- 180 degrees off, the complement: 1.10.
The first line is the one to sit with. Landing on an analogous neighbour, which is a real mistake, still scores 5.9 out of 10 on average and can score 9.65 if you happen to make it in the green band. Meanwhile the difference between a triadic miss and a full complementary reversal is 0.61 points. Past a certain distance, wrong is wrong and the scale stops distinguishing degrees of wrongness. Small hue errors are where all the recoverable score lives, which is the same conclusion the value-versus-hue measurement pointed at from the other direction.
You can feel this directly in a few rounds of solo play. Targets in the greens forgive sloppiness. Targets around cyan and azure punish the same size of error two or three times harder.
Using color theory without being fooled by it
None of the above means the wheel should be thrown out. It means using it with two corrections applied.
- Trust angles in the warm half, verify them in the cool half. From roughly hue 270 through red to hue 70, the standard wheel is within a few degrees of perceptually even. From 70 to 220 it is not, and a 30 degree move there can mean almost nothing or a great deal.
- Check the weakest pairing, not the average. A palette fails at its weakest link. Two colors in a four-color set that measure ΔE 25 apart will read as a mistake regardless of how far the other four pairings spread.
- Vary value before you vary hue. It is the larger lever and it works even when the hues are close, which is why a monochromatic palette can carry a whole layout while a badly built tetradic one falls apart.
- Prefer the artist complements when you need a safe pair. They average slightly less contrast than the geometric 180 degree pairs and they are far less likely to hand you a dud.
The short version
Color theory is a wheel, three properties, and six schemes built on fixed angles. The schemes work. The wheel is a mixing diagram that gets mistaken for a perception diagram, and measured with CIEDE2000 its twelve stations sit between ΔE 5.2 and ΔE 43.9 apart, with the green quarter compressed to about half its drawn size. Hue is only the second largest lever behind value. The artist wheel's complements average ΔE 83 against the screen wheel's 92 but vary half as much, which makes them the safer default. Past two colors every scheme averages around ΔE 68, so the choice between triadic, split complementary and tetradic is a choice about how weak you will let the weakest pairing get, and triadic wins that on a floor of 50.
Then go match a few colors and watch the theory meet a scoreboard.