Skip to content
Kolorvia
Color Theory

What Is Delta E? Measuring How Different Two Colors Really Are

Two colors can have very different RGB or HEX values yet look almost identical — or the reverse. Delta E is the standard way to measure perceptual color difference instead of relying on raw numeric distance.

· · 4 min read

On this page

Why you can't just compare RGB values

Subtracting RGB channels and calling the result "how different two colors are" doesn't match how people actually perceive color difference — RGB (and even HSL) distances don't scale evenly with perception. Two colors 20 units apart in RGB space might look nearly identical in one part of the color space and clearly distinct in another. Delta E was created specifically to fix this by measuring distance in a perceptually uniform space instead.

How it's actually calculated

Delta E is computed in CIELAB space — a color space built so that Euclidean distance between two points approximates perceived difference. The simplest version, CIE76 (the one this site's calculator uses), is literally the straight-line distance between two LAB coordinates: the square root of the sum of squared differences in L (lightness), a (green–red), and b (blue–yellow). It's fast to compute and accurate enough for the overwhelming majority of design and development use cases.

Reading the numbers

Under 1.0 is generally imperceptible to the human eye. Around 1–2.3 is a "just noticeable difference" — detectable only under close side-by-side comparison. From roughly 2–10, the difference is noticeable at a glance. Beyond that, colors read as clearly, unambiguously different. The 2.3 threshold specifically shows up a lot in printing and manufacturing tolerances, where it's a common pass/fail bar for "close enough to the target color."

Where it's actually used

Print shops and manufacturers use Delta E to quality-check that a printed or manufactured color matches a target within tolerance. Design systems use it to catch near-duplicate colors that should probably be merged into one token. It's also the basis for "nearest named color" features — given an arbitrary hex value, Delta E finds the closest color in a reference set far more accurately than comparing hue values alone, since it accounts for lightness and saturation differences too.

CIE76 vs. newer formulas

Later refinements — CIE94, and especially CIEDE2000 — correct for some known perceptual non-uniformities that CIE76 doesn't fully account for, particularly around saturated colors. They're more accurate but significantly more complex to compute. For most practical purposes outside of demanding color-matching industries like textiles, CIE76 is accurate enough and far simpler to reason about.

A worked example

Take #767676 and #7C7C7C — grays six units apart per channel. CIE76 puts them around ΔE ≈ 2.5: side by side you can just spot the seam; in isolation nobody could say which is which. Now take #0000FF and #4B0082 — over 150 units apart in RGB space, and unmistakably different colors. But the reverse also happens: large numeric RGB gaps inside saturated blue regions can score a smaller ΔE than tiny gaps near gray, which is the entire reason the metric exists. The calculator makes any of these comparisons one paste.

Picking a threshold for your use case

  • Merging design tokens: flag pairs under ΔE 2 — if users can't tell them apart, they're one token wearing two names.
  • Brand color QA: the classic manufacturing tolerance is ΔE ≤ 2.3 against the reference value.
  • Distinct UI states: aim well above ΔE 10 between states that must be distinguishable at a glance — and remember users won't see them side by side.
  • Nearest-named-color lookups: smallest ΔE wins; hue-only matching fails badly on light and dark variants.

Where Kolorvia uses it

Every color page's "nearest named color" is a ΔE search across the CSS named set. The palette pages run pairwise CIEDE2000 between swatches under five color-vision simulations to flag pairs that collapse for CVD users. And the calculator exposes the raw metric for your own comparisons.

What the number can't see

ΔE compares two colors as isolated patches under identical viewing conditions. Real interfaces violate that constantly: surrounding colors shift appearance (simultaneous contrast), tiny elements are harder to discriminate than large swatches, and screen brightness or night-mode filters move everything at once. A ΔE of 4 between two chart lines may be plenty at full-screen width and invisible at 2 px wide on a dim laptop. Treat ΔE as a rigorous lower bound on distinguishability, then sanity-check in context — and for anything carrying meaning, add redundant cues anyway.

When CIE76 isn't enough

CIEDE2000 layers hue-, chroma-, and lightness-dependent corrections onto the plain LAB distance, fixing CIE76's known optimism in saturated blues and pessimism near neutrals. The formula is dramatically more involved — a page of weighting functions instead of one square root — which is why general-purpose tools default to CIE76 and reserve CIEDE2000 for where tolerance genuinely matters: textile matching, print QA, and (on this site) the palette CVD safety report, where "how close do these look" is exactly the question being asked.

Related guides

Try it yourself

Put the theory into practice

Everything this guide explains is built into a free, live tool — try the math yourself.

We use cookies for site preferences and, where enabled, Google Analytics and Google AdSense — which may use cookies to personalize ads based on your visits here and elsewhere. See our Privacy Policy for details. Your color tool inputs are never sent to us regardless of your choice.