Delta-E Explained

One-sentence answer: Delta-E is the measured distance between two colors in a perceptual space, designed so that the number corresponds to how different they look rather than how different their raw values are.

What Delta-E measures

Delta-E — written ΔE — answers a question that sounds simple and is not: how different are these two colors? The delta is the Greek letter for change and the E comes from Empfindung, the German word for sensation, which tells you the intent. It was never meant to measure a difference in numbers. It was meant to measure a difference in perception.

That distinction is the whole subject. Two colors can be far apart in RGB and look almost the same, or close in RGB and look obviously different. RGB and HEX describe how a display should be driven; they say nothing useful about how far apart two colors appear to a person. Any honest color-difference metric has to start somewhere else.

Why it needs CIE Lab

Delta-E is calculated in CIE Lab, a space built from actual measurements of human vision rather than from the behaviour of a device. It has three axes: L* for lightness from black to white, a* running green to red, and b* running blue to yellow.

The point of Lab is that it is approximately perceptually uniform — a fixed step in any direction should produce roughly the same perceived change. Because of that, the distance between two points is a meaningful stand-in for how different the colors look. You cannot do this in RGB or HSL, where a step of ten means something completely different depending on where you start.

The word doing the work in that paragraph is approximately. Lab is uniform enough to be useful and not uniform enough to trust blindly, and every improvement to Delta-E since 1976 has been an attempt to patch the gap.

76, 94, and 2000

There are three formula generations still in circulation, and knowing which one a tool uses matters more than most people assume.

• ΔE*ab (CIE76) — plain Euclidean distance between two Lab points. Elegant, trivially fast, and wrong in predictable ways. It treats every region of Lab as equally uniform, which it is not.
• ΔE*94 (CIE94) — introduced weighting so that differences in highly saturated colors count for less, which fixed part of the problem and left the rest.
• ΔE00 (CIEDE2000) — the current standard, and what this site computes. It adds separate weighting functions for lightness, chroma, and hue, plus a rotation term that specifically corrects the blue-violet region where the earlier formulas were worst.

Because the formulas are scaled differently, a ΔE76 value and a ΔE00 value are not interchangeable. The same color pair typically scores lower under CIEDE2000 — often substantially so for saturated colors — which means quoting a tolerance without naming the formula is meaningless.

What CIEDE2000 fixes

Three corrections do most of the work:

• Lightness weighting. We are less sensitive to lightness differences at the very dark and very light ends than in the midtones. CIEDE2000 scales the lightness term accordingly, so a small shift in a near-black does not register as dramatically as the raw distance suggests.
• Chroma weighting. A fixed numeric difference between two vivid colors is far less visible than the same difference between two muted ones. CIE76 counted both equally and consistently overstated how different saturated pairs looked.
• The blue rotation term. The single most notorious failure of the older formula. Around the blue-violet region the interaction between chroma and hue is unusual, and CIE76 produced numbers that simply did not match what people reported seeing. CIEDE2000 adds a rotation term that applies specifically there.

This is why a deep navy and a slightly purple navy can score a comfortable ΔE76 while looking clearly different to a client — and why anyone matching blues on the older formula ends up mistrusting the metric. The formula was the problem, not their eyes.

Interpreting the number

These are the bands we use across the site, calculated with CIEDE2000 in CIE Lab under a D65 white point and the 2° standard observer:

• ΔE below 1 — imperceptible to the human eye in normal viewing.
• ΔE 1 to 2 — perceptible only on close inspection, typically when the two colors are directly adjacent.
• ΔE 2 to 3.5 — a slight difference, visible under direct comparison but easy to miss when the colors are apart.
• ΔE 3.5 to 5 — noticeable, though still acceptable for a lot of general production work.
• ΔE 5 to 10 — clearly different colors. Review the match rather than approving it.

Bands are a decision aid, not a policy. The right tolerance depends on what the color is doing: a corporate logo reproduced beside a printed reference deserves a tighter budget than a background tint nobody will ever see next to anything.

Production tolerances

Formal print standards put numbers on this, and it is worth knowing roughly where they land.

ISO 12647-2, which governs offset process printing, has traditionally specified a tolerance around ΔE*ab 5 for solid ink patches — but that figure is expressed in the older CIE76 formula. Research converting that tolerance into CIEDE2000 terms puts the equivalent in the region of ΔE00 3.3 to 3.5, which is why you will see modern shops working to roughly that number.

ISO 12647-7, covering contract proofing, moved to CIEDE2000 in its 2016 revision and applies tighter limits — measured spot colours are expected to fall within about ΔE00 2.5 of target. Proofs are held to a stricter standard than press sheets because a proof is the contractual reference everything else is judged against.

If you are writing a specification, name three things: the tolerance, the formula, and the measurement condition. “ΔE00 ≤ 2.5, measured M1, D50” is actionable. “Delta-E under 2” is not.

What Delta-E cannot tell you

The metric is genuinely useful and routinely over-trusted. It is blind to:

• Metamerism. Two colors matching under one light and diverging under another. A single ΔE figure calculated for one illuminant will never reveal it.
• Gloss, texture, and finish. A matte and a gloss sample can measure nearly identically and look obviously different, because surface behaviour is not part of the model.
• Metallics and fluorescents. Directional reflection and re-emitted light both break the underlying assumptions. Computed values here are labels, not predictions.
• Area and context. A difference invisible on a small swatch becomes obvious across a large solid, and a color's neighbours shift how it reads.
• Which direction the error runs. ΔE 3 too light and ΔE 3 too saturated are the same number and very different conversations with a client. The underlying Lab deltas tell you which; the summary figure does not.

Using it in a real workflow

1. Use it to shortlist, not to decide. Its best job is reducing hundreds of candidates to the three worth proofing.
2. Set the tolerance before you look. Deciding what is acceptable after seeing the number is how standards quietly slip.
3. Match the tolerance to the stakes. Primary brand colors, secondary palette, and incidental tints should not share one threshold.
4. Read the component deltas on anything critical. Knowing the error is mostly lightness rather than hue changes what you ask the printer to adjust.
5. Confirm physically. For anything that matters, the number narrows the field and a proof under controlled lighting settles it.

Every match on this site reports its CIEDE2000 score openly, including the bad ones. A converter that hides the distance is asking you to trust a match it will not quantify — our methodology page documents exactly how each score is produced.