Do colour harmony rules actually work?

Partly. The colour wheel rules, like complementary or triadic, are a good way to find candidate colours, so they're a fine place to start. But they don't predict which pairs will look good. When people are tested, they agree closely about which pairs feel harmonious, and those judgements depend on two things: which hues you start from, and how light each colour is. A rule that only measures the angle around the wheel can't see either of them.

Why does a palette that follows the rules still look wrong?

Say you picked a complementary palette in Canva, followed the rule exactly, and it looked awful. You didn't do it wrong. The rule did what it says. It just never promised what you were hoping for.

That's because every colour wheel rule makes the same hidden assumption, which is that all hues behave alike. "Complementary" means "go halfway round the wheel", and that only works as a rule if halfway round means the same thing whether you start at red or at blue. Two of the best studies on colour pairs, using different methods, both found that it doesn't.

The rules also ignore three things that shape how a pair looks: how light each colour is, how strong each colour is, and how much of each one you use. None of those is on the wheel.

I think of the colour wheel as a rhyming dictionary. Ask it for words that rhyme with "night" and it will give you two hundred, reliably and fast. What it won't tell you is which one belongs in your poem, and it has no opinion at all on whether the poem is any good.

The rules came from teachers, not experiments

Itten, Munsell, Chevreul and Ostwald each left a classic theory of colour harmony. In 2007 Stephen Westland and colleagues went back through all of them in a review for Colour: Design & Creativity, and found that every one had been criticised for lacking both a reason it should work and evidence that it does. None of them was complete.

The review also looked at the most famous attempt to turn harmony into a formula. Moon and Spencer published a mathematical harmony model in 1944, and in the tests since then it has performed quite poorly. The review's conclusion, which leans on the colour scientist Kuehni, is a plain one: there are no universal laws of colour harmony. What people call harmonious shifts with fashion, culture, generation and use.

That makes the rules old, tidy and untested. It doesn't make them useless. If you're curious how the wheel itself was built, that story is in Where the colour wheel came from.

Worth knowing: most popular articles skip the research. I read the five top-ranking pages on colour harmony, and four of them presented the schemes as settled fact. The fifth called the formulas unsound but cited no experiment. One says "studies show" without naming a study. And a widely shared magazine headline claims there's a scientific reason complementary colours look good, when the main study on that question found the opposite.

In 2011, same-hue pairs scored highest

Karen Schloss and Stephen Palmer set up their 2011 study carefully. They built 32 colours from eight hues, each at four mixes of lightness and strength. Then 48 people rated every pair, shown as a small square on a larger background, on a scale from "dissonant" to "harmonious". The study is free to read on PubMed Central (opens in a new tab).

People rated pairs as most harmonious when both colours shared the same hue, and the scores fell steadily as the hues moved apart. Complementary pairs, the ones straight across the wheel, were rated reliably less harmonious than the same hues paired with their near neighbours. That's why the authors concluded there was essentially no evidence for the old claim, going back to Chevreul, that contrasting hues are harmonious.

Are "harmonious" and "liked" the same thing?

No, and the same study shows it. Schloss and Palmer asked people both questions about the same pairs, and the answers behaved differently. Personal colour taste explained 21.7% of the differences in which pairs people liked, but only 1.4% of the differences in which pairs they called harmonious, which is roughly ten times less.

In other words, harmony is the more objective judgement of the two. People agree closely about whether a pair goes together, but they disagree much more about whether they like it. The two are still strongly linked, sharing about 62% of their variation. Related, then, but not identical.

A side note: you'll often read that colour harmony is just personal taste. The 2011 study points the other way. Individual taste barely moved people's harmony ratings, just 1.4% of the differences between them, but it moved their liking ratings far more, at 21.7%. So people do share a sense of which colours go together. They just don't share the same favourites.

When one colour sits on another, lightness takes over

Schloss and Palmer tested one more set-up in that 2011 study. They showed one colour as a shape on another colour as the background, and people preferred shapes on backgrounds of contrasting lightness, so light on dark, or dark on light. Lightness contrast alone explained 58.4% of the differences in those preferences.

That matters, because almost everything you design is one colour on top of another: text on a background, a button on a card. For that kind of pairing, the study's strongest finding isn't about hue at all. It's about how much lighter one colour is than the other. So if a pair isn't working, look at the lightness before you start changing the hue.

What did the 2026 study find, and why does it disagree?

It found the opposite about complementary colours. In June 2026 Ortensia Forni, Alexandre Darmon and Michael Benzaquen published a paper in iScience in which 346 people judged pairs of hues, and preference peaked when the two hues sat between 160 and 220 degrees apart. That's right around the complementary position. The paper is open access (opens in a new tab), so you can read it yourself.

Part of the story is that the methods were different. The 2011 study used a small square on a background and asked for a score. The 2026 study used checkerboards, with both colours covering equal areas, and asked people to pick the three most and three least harmonious pairs from a set of twelve. And where the 2011 study varied lightness and strength on purpose, the 2026 study held its picker's saturation and lightness numbers fixed, at 0.8 and 0.5, and changed only the hue.

So the two best datasets on this question disagree about complementary colours. The 2026 paper cites Palmer and Schloss on other topics, but not on their opposing harmony result, and I couldn't find anywhere the two are set side by side. I won't declare a winner. Both are careful studies, and complementary harmony is contested.

On the part that matters most, though, the two studies agree. They disagree about complementary colours, but both show that a rule which treats every hue the same can't work. In 2011, harmony depended on how similar the hues were. In 2026, it depended heavily on which hue you started from. Either way, "go 180 degrees" isn't enough information.

Some hues are easier to pair than others

In the 2026 study, blues (at 200, 220 and 240 degrees on the hue dial), yellow (60 degrees) and orange (40 degrees) combined well with most partners. Red (0 degrees), green (120) and purple (300) combined badly.

That's the finding that hurts the wheel rules most. The authors say plainly that the assumption behind the classic harmony models, that every hue behaves the same way, can't reasonably be upheld. How well a pair works depends a great deal on where you start.

The study also compared people's choices with the colours in 12,000 photos of natural landscapes, and checked those against two more sets of 4,319 and 15,501 images. The hues people combined most happily broadly matched the hues common in those scenes. But the authors describe that match by eye. They don't give a figure for how close it is, so I won't either.

Could brightness explain the 2026 result?

Possibly. I want to be careful here, though, because this part is my own analysis, not a finding of either paper, and no one has reviewed it yet. So here's what I noticed, with the working, and you can judge it for yourself.

Holding the picker's lightness at 0.5 doesn't hold brightness steady. I worked out how light each hue in that design actually looks, on a 0 to 100 scale built from how people judge lightness. Across the full set of 18 hues the study started from, spaced every 20 degrees, the answers ran from 31 for blue all the way to 89 for yellow. That's a spread of nearly 58 points, on colours the slider calls equal.

Then I checked every pair. In that set, the further apart two hues sit on the wheel, the bigger the difference in how light they look. Hues 20 degrees apart differ by about 7 points on average. Hues 180 degrees apart differ by about 33. The two measures rise together almost perfectly.

Which means that in that design, "far apart on the wheel" and "very different in brightness" came bundled together. Now, the 2011 study found that people like a big brightness difference when one colour sits on another. If people simply liked a big brightness difference here too, that alone could produce the peak the 2026 study reported. We can't tell whether they were choosing the hue gap or the brightness gap.

What brightness can't explain

There's one check that keeps the brightness idea honest, and it's this: brightness doesn't explain which hues combine well. The hues the 2026 study found combine best (three blues, yellow and orange) have an average lightness of 59.5 on my scale. The hues that combine worst (red, green and purple) average 61.5. They're as good as level.

That means the hue-specific result, that blues and yellows pair well while reds, greens and purples pair badly, isn't a brightness side effect. It stands on its own. The authors also knew their picker's numbers don't match how colours look, and they dropped 5 of their original 18 hues for that reason. That reduced the problem, but it didn't remove it.

My reading is a narrow one. Brightness difference may account for the complementary peak. It doesn't touch the bigger point both studies share, though: the starting hue matters, and an angle on a wheel can't capture it. If a colour scientist reads this and finds my working wrong, I'll correct it here. And if you want to know why two colours at "the same lightness" can look so different, that's explained properly in Why "same brightness" colours don't look the same.

Do harmonious colours have any measurable benefit?

Yes, but the benefit comes from similar colours, not from harmony as such. Thomas Sanocki and Noah Sulman tested it, and their study came out in Perception in 2011. Palettes of four colours with similar hues were rated more pleasant than mixed palettes, and people could hold more of them in short-term memory too, remembering 26% to 45% more coloured squares.

Then they ran a final experiment to pull "similar" and "harmonious" apart, and similarity won. The advantage came from the colours being alike, not from them being harmonious. The titles show the shift. An early conference summary of the work, from 2009, had "harmony" in its title. The published paper is called "Color Relations Increase the Capacity of Visual Short-Term Memory", and its conclusion reverses the old title.

For your design, the useful lesson is that related hues are easier to take in at a glance.

Why does every tool give a different "complementary" colour?

Because they're using different wheels. As Westland's 2007 review points out, "complementary" has at least three technical meanings: one from mixing paint, one from mixing light, and one from the afterimage you see after staring at a colour. Design software and textbooks flatten all three into "the opposite side of the wheel".

On a paint wheel, the opposite of red is green. On a light wheel, the opposite of red is cyan. Adobe says its Adobe Color harmonies are worked out on the paint wheel, known as RYB, even though the colour codes it shows you are screen colours. An Adobe employee confirmed this in Adobe's own community forum (opens in a new tab) in March 2023, and that's why its complement of red isn't the number you'd expect.

So when Canva, Figma and Adobe each hand back a different complementary colour, none of them is broken. They're just answering on different wheels. Paint and stage lighting really do follow different rules: mix red and green paint and you get mud, but red and green light together make yellow. Each process has its own wheel, with its own opposites.

People don't pick the mathematical opposite either

Not even on a wheel built to match human vision. Forough Mahyar and colleagues tested it in 2007, in a study published in the AIC 2007 conference proceedings. They showed 10 people with normal colour vision 20 reference hues, and for each one, the person dialled in the hue that gave the most contrast, with lightness and strength held fixed.

What they chose wasn't the opposite hue on that wheel. The gaps were large in the yellow-greens and the blues. And the choices weren't reversible. People who picked a blue as the strongest contrast for a pinkish red then picked a yellowish orange, not red, as the strongest contrast for that same blue. The authors call this hard to explain, and they suggest the cause may be language rather than vision.

Ten observers is a small group, so don't lean on it too hard. But it does mean "the opposite colour" now fails four ways: on paint wheels, on light wheels, on wheels built to match vision, and in what people actually choose.

Do complementary colours make text easy to read?

No. Being opposite on the wheel says nothing about brightness, and brightness difference is what makes text readable. I checked six fully saturated complementary pairs, with both colours at the same lightness number, and four of the six failed the web's accessibility minimum for body text, which is a contrast of 4.5 to 1.

Same rule, same strength, same lightness number, and yet the results run from 1.52 to 8.00. Orange and azure, a textbook complementary pair, barely differ in brightness at all.

One hue can actually beat two. A single blue at two lightnesses (hue 210, fully saturated, at lightness 0.15 and 0.85) reaches 10.32 to 1, which is more than three times the red and cyan pair. So the scheme with the least hue variety wins on the thing that matters most for text. If you want more on contrast ratios themselves, have a look at Can everyone actually read your design?

What should you do differently in your next palette?

Use the wheel to find candidates, and then judge each pair on what you can actually see.

1. Treat the scheme as a shortlist, not a verdict. Let the wheel suggest five options. Then look at each pair yourself, at real size, on a real layout.
2. Check brightness before hue. For anything that sits on top of something else, like text, icons or buttons, make sure one colour is clearly lighter than the other. Hue opposition won't do that for you.
3. Test contrast with numbers. Run every text and background pair through a contrast checker. Don't assume a complementary pair passes, because most in my test didn't.
4. Don't trust "same lightness" across hues. Two colours at lightness 50 in your picker can look very different in brightness. Compare them side by side, or in an eye-weighted greyscale.
5. Expect your tools to disagree. If two apps give you different complements, they're probably using different wheels. Pick the one that looks right in your design, not the one that's "correct".
6. Try one hue first. For text and backgrounds, a single hue at two very different lightnesses is often more readable, and calmer, than a complementary pair.
7. Decide how much of each colour, too. The wheel says nothing about proportion, and how much of each colour you use changes how a pair reads. Building a brand palette that works covers giving colours roles.

Frequently asked questions

Why do some colour combinations look great when they break every rule?

Because the rules were never predictions. Colour wheel schemes generate candidates, and they assume every hue behaves the same way, which the experiments don't support. A 2026 study found that how well a pair works depends heavily on the starting hue. Hue angle is also just one of at least four things doing the work. Lightness, colour strength and how much of each colour you use matter too, and none of them is on the wheel.

Why does every tool give me a different "complementary" colour?

Because each tool is using a different wheel. On a paint wheel, red's opposite is green, whereas on a light wheel, the kind screens use, it's cyan. Adobe says Adobe Color works out its harmonies on the paint wheel while showing you screen colour codes. So there's no single true complement to look up. In one study, people asked to pick the strongest contrast for a pinkish red by eye chose a blue.

Do complementary colours always give you strong contrast?

No. Complementary means far apart in hue, not far apart in brightness, and brightness is what makes text readable. When I tested six fully saturated complementary pairs at the same lightness setting, four failed the web's 4.5:1 minimum for body text. Red on cyan reached only 3.19:1, while blue on yellow reached 8.00:1. A single blue at two lightnesses reached 10.32:1, beating every complementary pair.

Can a palette follow the rules and still look boring?

Yes, because harmony and liking are different judgements. In a 2011 study, personal taste explained 21.7% of the differences in which colour pairs people liked, but only 1.4% of the differences in which pairs they called harmonious. So people largely agree on what goes together, yet they still disagree on what they enjoy. That's how a palette can be genuinely harmonious and still feel flat.

The wheel tells me what goes with my colour. How do I choose the first one?

The wheel can't help you with that, because every harmony rule starts from a colour you already have. The best recent evidence does offer a lead, though. In a 2026 study, blues, yellow and orange combined well with most other hues, while red, green and purple combined badly. The hues people paired most happily broadly matched hues common in natural landscapes, but the study gives no figure for how close that match is.

For the curious

Schloss and Palmer separated harmony, preference and similarity. Their complementary result was F(1,47) = 17.67, p < .001, and harmony and preference correlated at r = +0.79. The equal-lightness problem is perceptual non-uniformity in HSL. My lightness figures are CIE L* from the sRGB values, using WCAG relative luminance. Across the 18 hues at HSL S 0.8, L 0.5, L* ran from 30.9 (240°) to 88.6 (60°). Averaged at each hue distance in that full 18-hue grid, mean |ΔL*| tracked hue distance at r = +0.989; across the 153 single pairs, r = +0.535. That makes lightness contrast a possible confound for the 2026 peak. It's a hypothesis, not a result. The contrast figures use the WCAG 2.x ratio. Mahyar and colleagues worked in CIELAB, at L* 52 and C* 30, and found deviations over 10 units at 90–150° and 260–300°. The 1944 model is Moon–Spencer. Westland's three complements are subtractive, optical (additive) and afterimage.

Sources (6)

• Karen B. Schloss and Stephen E. Palmer, "Aesthetic response to color combinations: preference, harmony, and similarity" (opens in a new tab), Attention, Perception & Psychophysics 73(3), 551–571, 2011
• Ortensia Forni, Alexandre Darmon and Michael Benzaquen, "Harmonious color pairings: insights from human preference and natural hue statistics" (opens in a new tab), iScience 29(6), art. 116038, 8 June 2026, CC BY
• Stephen Westland, Kevin Laycock, Vien Cheung, Phil Henry and Forough Mahyar, "Colour Harmony" (opens in a new tab), Colour: Design & Creativity 1(1), 1–15, 2007
• Thomas Sanocki and Noah P. Sulman, "Color Relations Increase the Capacity of Visual Short-Term Memory" (opens in a new tab), Perception 40(6), 635–648, 2011
• Forough Mahyar, Vien Cheung, Stephen Westland and Phil Henry, "Investigation of complementary colour harmony in CIELAB colour space" (opens in a new tab), AIC 2007, 82–85
• Adobe Community, "How is complementary color determined?" (opens in a new tab), Adobe Employee answer, 9 March 2023