How different do two colours need to be?

There isn't one number that tells you when two colours look different. What you can see depends on which colour it is, which way it moved, and how you're looking at it. The famous "under 1 is invisible" rule is just one developer's own guess, and the experts' estimates differ fifteen times over. The test that actually works is a simple one: put both colours side by side, touching, and look.

What the difference number measures

The E in Delta E comes from Empfindung, the German word for "sensation". That's a good clue to what it's for. Delta E, written ΔE, is a single number that estimates how different two colours look to a person.

The first popular version came out in 1976, and it's simple. Imagine every colour placed on a three-dimensional map, with how light it is running up the middle. The 1976 version just measures the straight-line distance between two colours on that map.

The thing is, the map was meant to be evenly spaced, so a centimetre would mean the same thing everywhere. It isn't, especially among the vivid colours. So better versions followed: one in 1984, one in 1994, and then the current standard from 2000, which the international colour body published in 2001. Each one bends the distance to fit what people actually see.

What most explainers skip is what the formula was built for. Printers, dye houses, paint makers and monitor calibrators use it to pass or fail work against a contract. They measure a sample, check it against the agreed number, and either ship it or redo it. It was never built to help you choose colours. So if you've been using a Delta E number to decide whether two colours in your palette are different enough, you've been using a quality-control tool for a design job.

Where does the "under 1 is invisible" table come from?

You've probably seen the table. 1 or less is "not perceptible by human eyes", 1 to 2 is "perceptible through close observation", and 2 to 10 is "perceptible at a glance". It sits on monitor makers' sites and instrument makers' blogs, usually stated as plain fact. But it all comes from one personal web page.

That page is called Delta E 101, and a developer named Zachary Schuessler wrote it. Directly under the table he says it was built from his own tests and "does not come from an authority source", and he tells readers to do their own testing for their own use. Just above the table, he even warns that two colours can look different when the number is below 1.

The page itself is honest. The problem is the copying. In September 2026 I read the ten top-ranking pages on the topic, and three of them reproduce the table, two word for word and one reworded. Only one, ViewSonic's, credits Schuessler. And none of them passes on his warning. In other words, the most copied fact about colour difference on the internet is one its own author openly disowned.

The bottom rows get copied too: 11 to 49 is "more similar than opposite", and 100 is "exact opposite". Those rows don't mean much, and he doesn't defend them. A score of 100 isn't even the top of the scale. Black against white happens to score exactly 100 on the 2000 version, but some pairs score more.

The experts can't agree on a threshold

Published estimates of the smallest difference people can see, on the 1976 version, run all the way from 0.2 to 3. In 2013 Thomas, Colantoni and Trémeau laid them side by side in a review:

That's a fifteen-fold range. The review's authors settled on 1.5 to 2 as the best compromise. I'm reporting these five figures as that review gives them, because I haven't read all five original papers myself.

Now it gets stranger. The CSS Color specification, which sets the web's own colour rules, says one "just noticeable difference" is 2 on the 2000 version. But the documentation for Color.js, a widely used colour code library, says that "for most DeltaE algorithms, 2.3 is considered the 'Just Noticeable Difference'". Both were written by the same two people, Lea Verou and Chris Lilley. If the people who wrote the rules give two different numbers, the number isn't settled.

Even 2.3 gets quoted two ways. The 2013 review reports it as the smallest visible difference being 2.3 units. But a 2011 survey by Mokrzycki and Tatol says the reverse, that one unit equals 2.3 just-noticeable steps. They can't both be right.

And the companies selling you screens don't agree either. EIZO says 1 to 2 can't be seen. BenQ says under 1 for experts and under 3 for anyone. A printing trade-body article puts a trained eye at 2 to 3 and an untrained eye at 5 to 6. None of them gives a source.

Why do two Delta E numbers for the same colours disagree?

Usually it's because they come from different versions of the formula. Take two vivid blues, #0000FF and #1A1AE6. On the 1976 version they're 16.6 apart, but on the 2000 version they're only 2.6 apart. Same two colours, and the numbers are more than six times apart.

And the gap isn't steady. Across twelve everyday pairs I checked, the 1976 number ran from about the same as the 2000 number (for mid greys) to more than six times bigger (for vivid blues). ISO's own colour proofing standard says plainly that "there is no simple correlation between the two metrics", and it warns that work which passed under the old formula may fail under the new one.

That would be fine if everyone wrote the formula down. Almost nobody does. Every monitor page I found that quotes a Delta E number names no formula, and the documents that do name one, like EIZO's and BenQ's calibration report guides, don't promise any particular number. That leaves you with a Delta E number and nothing to compare it with.

Even the reference code can trip you up. Color.js, written by the editors of the web's colour rules, still uses the 1976 version unless you ask for another, because it's faster and the library can't know which one suits you. The 2000 version is also fiddly to get right. Sharma and colleagues published a set of test colours in 2005 for exactly that reason: they'd found faulty versions of the formula online, some of them from reputable sources.

A side note: two correct tools can still give two different answers. Color.js and Coloraide, two well-known colour libraries, both match Sharma's published test results exactly. Yet for the same pair of blues, one gives 2.48 and the other gives 2.64, because each one quietly assumes a different "white" for the maths. In about 20,000 random pairs I tested, a quarter differed by more than 5%. Vivid blues differ the most, and greys don't differ at all.

One unit is a different size for every colour

This is the heart of it. I counted every one of the 16,777,216 colours a hex code can name, and checked how many sit within 1 unit of a starting colour on the 2000 version.

2,575 hex codes are within one unit of royal blue. Only 34 are within one unit of mid grey. That's a 75-fold difference in what "one unit" covers, and the 2000 version was designed specifically to even this out.

You've probably felt this already. Two greys that looked identical in your file turn out to be visibly different side by side, and two blues you thought were different turn out to match. Greys are fussy. Vivid blues are forgiving. The numbers just put a size on it, which means your greys deserve a much closer look than your blues.

None of this is new. In 1942 MacAdam showed that the smallest difference you can detect changes size across the colour map, and that no simple redrawing of the map could turn those regions into equal circles. Even on the later map built to fix the problem, his regions are still up to three times bigger in some places than in others.

Worth knowing: MacAdam's famous ellipses were drawn at ten times their real size. The 1942 picture of colour "ellipses" you'll find in textbooks is real, but read the small print. It comes from one observer making more than 25,000 colour matches through an artificial pupil, so it measured how consistently he matched colours, not the smallest difference he could see. And MacAdam drew every ellipse at ten times its true size.

Does it matter which way a colour moves?

It matters as much as how far it moves. A colour can change in three ways: it can get lighter or darker, it can get more or less colourful, or its hue can turn towards another colour. I worked out how far royal blue (#1D4ED8) has to move in each of those directions to reach one unit on the 2000 version.

One degree of hue uses up the whole budget. Whereas you can drain more than four times as much colourfulness out of that blue for the same number. In other words, your eyes are far pickier about which colour it is than about how much colour it has.

For a near-grey teal (#008080) the pattern flips. Losing 2.2 points of colourfulness is one unit, but the hue has to turn nearly 3 degrees. And a pure grey has no hue or colourfulness to change at all. It can only get lighter or darker, which is why its neighbourhood is so small.

That's why "how different?" is really three questions, and a single threshold can't answer all three. If you're matching a brand blue, watch the hue before anything else.

The conditions the formula was built for

Picture a brand colour on a phone in a café. Now compare that with what the international standard for the 2000 version says the formula is meant for. The right-hand column is my own illustration, not part of the standard.

The same standard flags a separate report from the international colour body on the formula's limits for differences under 2 units. That's exactly the range every threshold table argues about. Luo, who helped build the 2000 version, makes a related point: colour difference formulas only hold under the conditions that body defines, and they were made for two large patches, not for pictures.

It's a bit like a car's fuel economy figure, measured on a rolling road at a steady speed. It's real and it's useful, but it's not what your car does on the school run.

Two of those conditions are ones you can actually set up yourself: large patches, and the two colours touching. Both make differences easier to see, not harder. If you can't see a difference with big patches touching, it's unlikely anyone will see it on a small button.

The standard also assumes normal colour vision, and so does every figure on this page. How colours separate for people who see colour differently is covered in Designing for colour blindness.

Was "1 means just noticeable" ever true?

Not for the formula everyone quotes. The idea that 1 unit equals the smallest visible difference was only built in on purpose in 2019, when the ITU, the international television body, published a new difference measure for TV pictures. It says that on its scale, "a value of 1 indicates the potential of a just noticeable colour difference". That's the first standard I found that makes the promise in writing.

Even then there are limits. The promise only holds when your eyes are at their most sensitive, so the standard says it may overstate how visible a difference is. And for an actual reference screen, it suggests a working tolerance of under 3, not under 1. The new measure exists because the 2000 version was judged not good enough for modern wide-colour, high-brightness television.

In other words, the people who most needed "1 = just noticeable" to be true had to write a new formula to make it true, and then they told users to allow 3 anyway.

Industries pick their own numbers

Each industry sets its own tolerance, on purpose, for its own job. Here are a few, with their formulas, because a number without one doesn't tell you much.

Be careful with the Pantone figure, though. It's the goal for printing Pantone's own guides, not a tolerance for your job matching a Pantone colour, and Pantone itself says its guides are a starting point.

The proofing standard adds a striking note. Tolerances tighter than 2.5 aren't practical yet, it says, because different measuring instruments don't agree well enough. Which means two instruments can disagree by more than the difference you're arguing about. Why print can't hold some screen colours at all is covered in Why bright designs go dull in print.

One industry can even run three thresholds at once. Fogra, a print research institute, grades the same measurements into three quality bands: an average under 2.5, 4.5 or 5.5 on the 2000 version. Textile fading grades have been written as units of difference since 1993 at least, where grade 4 is 1.7 and grade 3 is 3.4. And X-Rite says car makers often work to 0.5 or less, on yet another formula.

How far apart are the colours you actually compare?

Most everyday pairs are nowhere near the line, but a few sit right on it. I measured the pairs designers fuss over most, all on the 2000 version.

White against #FAFAFA is 1.00, to two decimal places. White against #F5F5F5 is 2.01. The two commonest "is this grey even different?" choices in interface design land exactly on the two most quoted thresholds. The whole range from #F5F5F5, a standard card background, up to pure white is only two units wide.

Greys are touchy at the dark end too. On #121212, changing the green part of the hex code by one, which is the smallest edit there is, already makes a whole unit of difference. On brand yellow #FFCC00, though,, nudging the red part by one does nothing, because it's already at the maximum.

The good news is at the other end. Neighbouring steps of a published colour scale run 5 to 11 units apart, so if you're building a palette, you're nowhere near the threshold. Material's dark-mode panels, at 3 to 5 units apart, sit above every published threshold but below anything that reads as a new colour. That looks deliberate.

Brand colour drift sits somewhere in between. Yellow #FFCC00 against #FFC800 is 1.36, which is inside Pantone's own goal. Crimson #DC143C against #E11D48 is 2.85, which is outside it.

If the same hex code looks different on your phone and your laptop, that's a screen problem rather than a difference problem, and it's covered in Why your brand colour looks different on your phone. And if a colour looks different on a different background, have a look at The same colour looks different in different places.

A big difference isn't the same as good contrast

A red (#D6362C) and a green (#2E8B3A) that are equally light sit 65 units apart on the 2000 version. They obviously look different. But their contrast ratio is 1.10:1, so text in one on the other is unreadable. That's because colour difference and contrast answer different questions. As a rule of thumb, strong contrast always means a big colour difference, but a big colour difference tells you nothing about contrast. How readability is measured is covered in Can everyone actually read your design?

What should you do differently next time you compare two colours?

Use the number to rank pairs, not to rule on them. Then decide by looking.

1. Put the two colours side by side, touching. Make the patches big. Touching edges and large patches are the conditions where differences are easiest to see, so if you can't see it there, it's unlikely anyone will see it on a small button.
2. Check on the screen it will be seen on. Use a phone if it's for phones, and look at it in the light where it will be used.
3. Trust your eye over the table. The "under 1 is invisible" rule is one person's guess, and he said so himself.
4. Be fussier with greys than with vivid colours. A grey has far less room to move before someone notices.
5. Watch the hue most. When you're matching a brand colour, a tiny turn of hue shows sooner than a loss of colourfulness, especially in blues.
6. If you must use a number, use the 2000 version and say so. Use it to sort pairs from closest to furthest apart, and then go and look at the closest ones yourself.
7. Ask any supplier which formula they mean. A printer, a monitor spec or a brand guide that says "under 2" without naming a formula hasn't told you much.
8. Don't panic if two tools disagree. Different formulas and quiet assumptions can move the number by a lot, and neither tool is necessarily wrong.

Frequently asked questions

What Delta E is visible to the human eye?

There's no agreed number. Published estimates of the smallest visible difference range from 0.2 to 3, and a 2013 review that compared them settled on 1.5 to 2 as the best compromise. The popular table saying "1 or less is not perceptible" comes from a personal web page whose author says it "does not come from an authority source". How visible a difference is also depends on which colour it is and which way it moved.

What is a good Delta E value for a monitor?

Under 2 is the figure most monitor makers quote, but a spec sheet that gives a Delta E number almost never says which formula produced it, and the formulas can differ by several times on the same pair of colours. According to ISO's proofing standard, numbers below about 2.5 are also near the limit of what two measuring instruments agree on. So treat a monitor's Delta E claim as a rough grade, not a measurement.

What Delta E is acceptable for printing?

It depends on the standard and the part of the job. ISO's offset printing standard, in its 2004 edition, allows solid inks up to 5 units from the target on the 1976 formula. ISO's proofing standard, which switched to the 2000 formula in 2016, allows 3.0 on solids and an average of 2.5. Pantone aims for under 2 on its own guide printing. Ask your printer which standard and formula they use.

What's the difference between Delta E 76 and Delta E 2000?

The 1976 version measures the straight-line distance between two colours on a colour map that turned out to be unevenly spaced, so it exaggerates differences among vivid blues. The 2000 version corrects for that, and it's the current standard. You can't swap one for the other: two vivid blues score 16.6 on the 1976 formula and 2.6 on the 2000 one. ISO's proofing standard says there's no simple correlation between them.

What does a colour deviation of 4.65 in Delta E 2000 look like?

There's no fixed look, because the same number covers different amounts of change for different colours. Around a vivid blue, about 134,000 hex codes sit within 5 units. Around a mid grey, only about 5,000 do. For scale, Material Design's dark-mode panel greys sit 3 to 5 units apart, which is enough to read as a step but not as a new colour. The only reliable answer is to put both colours side by side and look.

For the curious

The difference number is ΔE. The 1976 version is CIE76 (ΔE*ab), which is straight-line distance in CIELAB; then came CMC (1984), CIE94, and CIEDE2000 (ΔE00), standardised as ISO/CIE 11664-6. My figures use CIEDE2000 with k_L = k_C = k_H = 1, and CIELAB under a D65 white, cross-checked in Color.js 0.7.1, Coloraide 8.12.1 and my own code against Sharma's 34 test pairs. The library disagreement comes down to D50 against D65 CIELAB. "Colourfulness" here is CIELAB chroma (C*). The TV measure is ΔE_ITP (ITU-R BT.2124). MacAdam's criterion was the standard deviation of colour matching, under Illuminant C. The CIE report flagged by the standard is CIE 230:2019, and I haven't read it. CSS Color 4 puts one step at ΔEOK 0.02; in my test, pairs at that distance had a median CIEDE2000 of 2.4, with 90% between 1.6 and 4.9. Oklab itself was tuned using CIEDE2000, so it isn't an independent check. And one number changed form between documents: Dolby's ICtCp paper credits a "ΔE2000 of 3.0" to the DCI specification, which only states its white-point tolerance as a chromaticity.

Sources (17)

• ISO/CIE, ISO/CIE 11664-6:2022, Colorimetry, Part 6: CIEDE2000 colour-difference formula (opens in a new tab), free preview, Clause 5
• Sharma, Wu and Dalal, "The CIEDE2000 Color-Difference Formula: Implementation Notes, Supplementary Test Data, and Mathematical Observations" (opens in a new tab), Color Research & Application 30(1), 21–30, 2005
• Thomas, Colantoni and Trémeau, "On the Uniform Sampling of CIELAB Color Space and the Number of Discernible Colors" (opens in a new tab), CCIW 2013, Springer LNCS
• MacAdam, "Visual Sensitivities to Color Differences in Daylight" (opens in a new tab), JOSA 32(5), 247–274, 1942
• Luo, "Development of colour-difference formulae" (opens in a new tab), Review of Progress in Coloration 32, 2002
• W3C, CSS Color Module Level 4 (opens in a new tab), §14.2.1 and §20, Candidate Recommendation Draft, 13 September 2026
• Lea Verou and Chris Lilley, "Color differences" (opens in a new tab), Color.js documentation
• ITU-R, Recommendation BT.2124-0 (opens in a new tab), January 2019
• ISO, ISO 12647-7:2016 (opens in a new tab), proofing, free preview; ISO 12647-2:2004, offset printing, Table 3
• Pantone, "Normal Printing Process Color Variances, Explained" (opens in a new tab)
• 3M, "Corporate General Specification RD1200" (opens in a new tab), 30 March 2018
• Fogra, PSD Handbook 2025 (opens in a new tab), and ISO, ISO 105-A02:1993 (opens in a new tab), free preview
• X-Rite, "Defining a Realistic Pass/Fail Tolerance" (opens in a new tab)
• Dolby Laboratories, "What is ICtCp?" (opens in a new tab), white paper v7.1, 2016
• Björn Ottosson, "A perceptual color space for image processing" (opens in a new tab), 23 December 2020
• Shelby Sapusek, "Dissecting Delta E and the mathematical difference between colors" (opens in a new tab), Printing United Alliance, 18 February 2025
• Credited in plain text, not linked: Zachary Schuessler, Delta E 101; Mokrzycki and Tatol, "Colour difference ΔE: A survey", Machine Graphics and Vision, 2011; EIZO and BenQ calibration report guides