Perception
Colour Discrimination Test
Six squares, one slightly different. Find it as the difference shrinks. You get three thresholds rather than one — lightness, green–red and blue–yellow — because your sensitivity to those differs by more than twice, and a single number would hide it.
Colour discrimination
Six squares appear; five are identical and one is slightly different. Click the odd one. Each time you are right the difference gets smaller. The test runs three times — once for lightness, once for green–red, once for blue–yellow — because your sensitivity to those is not the same, and reporting one number for all three would hide that.
- Set your screen brightness where you normally use it and don't change it mid-run.
- Avoid glare and direct sunlight on the screen. Ambient light washing across a panel raises the threshold and we cannot detect it.
- Turn off night mode / blue-light filters. They shift colour deliberately, and the blue–yellow ladder in particular will measure the filter rather than you.
What your equipment contributed
- Smallest step we can render
- 0.38–0.56 ΔEone 8-bit code value, by axis
- Finest step offered
- 1.0 ΔEnever finer than the code step
- Panel, brightness and room light
- unmeasurableand they dominate the result
Colours here are written as 8-bit sRGB values, so there is a smallest step the page can express — ask for a difference finer than one code value and the browser renders the colour it was already showing. We measure that step for each axis and refuse to ladder below it, which is why the finest difference offered is what it is rather than an arbitrarily small number. The larger limits are the ones we cannot measure: your panel, its calibration and brightness, and the light in your room. An uncalibrated display can be off by far more than the thresholds being measured here, and glare across a screen raises them further. So this is a reading taken on your equipment in your room — it is not a measure of your colour vision in the clinical sense, and a different screen would give you a different number.
How this is measured
What the difference is measured in
ΔE*ab — straight-line distance in the CIELAB colour space, which is designed so that equal distances correspond roughly to equal perceived differences. Because each ladder moves along a single Lab axis at a time, the ΔE of a step is simply the size of that step, and it is the same quantity the research below reports per axis. That is why we vary one axis at a time rather than wandering diagonally: it keeps our number and the published number the same kind of thing.
The ladder, and how an axis ends
Each axis starts at a difference of 8 ΔE — obvious to almost anyone — and steps down through 6, 4.5, 3.5, 2.5, 1.8, 1.3 and 1.0. Get it right and the next difference is smaller. Two misses at the same step ends that axis, and your threshold is the last step you cleared; one miss is weak evidence when a blind guess succeeds one time in six. Missing the very first step twice reports no threshold rather than a score, because nothing was ever cleared.
Those step sizes are what we aim for. The figure we report back is what the step actually turned out to be once both colours were rounded to whole 8-bit values — see below, because at the fine end of the ladder the difference between the two is not small.
Why the ladder stops where it does
Colours on a web page are written as 8-bit sRGB values, so there is a smallest step the page can express — roughly 0.4 ΔE on lightness and green–red, and around 0.56 on blue–yellow. Ask for a finer difference and the browser renders the colour it was already showing. We measure that step for each axis and refuse to ladder below it. A test that carried on past that point would be asking you to distinguish two patches made of identical pixels, then recording the failure as a fact about your eyes.
The same rounding is also why we report the difference that was rendered rather than the one we asked for. Both colours snap to whole code values, so a requested 1.0 ΔE actually arrives somewhere between about 0.68 and 1.15 depending on the axis and which direction the odd square was offset — a 15 to 30% error at the fine end of the ladder. It is negligible at 8 ΔE and it is not negligible at 1. Printing the number we aimed at would be claiming a precision the squares on your screen never had.
The squares, and what they sit on
Six squares, five identical and one offset along the current axis in a randomly chosen direction — so the odd square is not reliably the lighter or the redder one, which would be learnable within a run. They sit on a uniform darker grey field rather than directly on the page, because the immediate surround of a colour affects how it looks, and leaving that uncontrolled would let the design of the page influence the measurement.
What we cannot tell
Whether you guessed. With six squares a blind guess is right one time in six, and the two-attempt rule means roughly a 30% chance of clearing a step you could not actually see — so a threshold may sit one step finer than the truth. We also cannot see your panel's calibration, its brightness, your viewing angle, whether a night-mode filter is shifting everything, or how much light is falling on the screen. Any one of those moves the result by more than the steps being measured. Nothing is timed here, so no display refresh or input latency enters into it, and nothing you do is stored or transmitted.
How your result compares
Colour perceptibility thresholds have been measured on screens, so a reference does exist — with the same kind of caveat as the rest of this site. In a study of 16 observers with normal colour vision, viewing computer-generated stimuli on a spectrally calibrated monitor, the 50% detection thresholds came out markedly different by axis:
| Axis | 50% detection | Signal detection (d′ = 1) |
|---|---|---|
| Lightness (L*) | 1.0 | 1.25 |
| Green–red (a*) | 1.0 | 1.25 |
| Blue–yellow (b*) | 2.6 | 2.8 |
The useful part is the ordering and the ratio: blue–yellow sits roughly two and a half times coarser than the other two. If your own three numbers show that same shape, your colour vision is behaving the way the literature says it should — regardless of whether the absolute values line up.
What we matched, and what we could not
Matched: the units (ΔE in CIELAB), and the decision to measure each axis separately rather than pooling them.
Not matched: their monitor was spectrally calibrated every week and viewed from a fixed 1.2 m in controlled conditions; yours is whatever you have, wherever you are sitting. Their stimuli were small tooth-shaped patches under half a degree across, shown for at most two seconds — ours are large squares you can study for as long as you like, and bigger, unhurried stimuli are easier, so our thresholds should come out finer than theirs for that reason alone. Their task compared a pair; ours is an odd-one-out among six. Their observers were dental professionals and patients judging simulated teeth, n=16.
So read the table as the expected shape of sensitivity across the three axes, not as a line your numbers should land on. A commonly repeated rule of thumb puts the just-noticeable difference at about 2.3 ΔE overall; it comes from a different task under different conditions, and the axis-by-axis figures above show why a single overall number was never going to be the whole story.
No percentile is shown. The source gives threshold estimates for sixteen observers under calibrated conditions, not a distribution we could place you within — and the gap between a calibrated laboratory monitor and an unknown consumer panel is wider than the thresholds themselves. A percentile computed from BlinkBench's own visitors would be worse: it would rank displays and room lighting at least as much as it ranked eyes.
Source: Lindsey, D. T., & Wee, A. G. (2007). Perceptibility and acceptability of CIELAB color differences in computer-simulated teeth. Journal of Dentistry, 35(7), 593–599. doi:10.1016/j.jdent.2007.03.006
Last reviewed: July 2026
Frequently asked questions
Why does this give three numbers instead of one?
Because your eye is not equally sensitive in every direction through colour space, and averaging over that difference throws away the most interesting thing the measurement found. The research this page cites measured a threshold around 1.0 ΔE for lightness and green–red but roughly 2.6 ΔE for blue–yellow — a difference of about two and a half times. A test that wanders in random directions and reports a single figure is reporting an average of those, weighted by whichever directions it happened to pick. Three ladders take longer to run. They are also the only version of this that means anything.
Why is my blue–yellow number so much worse than the other two?
Because that is how human colour vision works, and your result is behaving normally. The retina has far fewer short-wavelength (blue) cones than medium and long wavelength ones, and they are almost absent from the very centre of the fovea. Published thresholds on the b* axis run roughly two to three times higher than on L* or a* for exactly this reason. Your display contributes too — the blue–yellow direction also has the coarsest rendering step of the three on an 8-bit screen, which we measure and show you on the results panel.
Does this test for colour blindness?
No. Colour vision deficiency is diagnosed with plates or arrangement tests designed to separate specific confusion lines — Ishihara, Farnsworth–Munsell and similar — under controlled lighting. This measures a detection threshold along three axes on an uncalibrated screen. Someone with a red–green deficiency would likely show a notably worse green–red number here, but a poor result is far more likely to mean your screen, your brightness or your room light than anything about your eyes, and a good result rules nothing out. If you have any reason to wonder about your colour vision, see an optometrist.
Would I get a different result on a different screen?
Almost certainly, and that is the honest headline. An uncalibrated panel can be off by considerably more than the thresholds being measured here, and brightness, viewing angle, glare and ambient light all move the number. Night-mode and blue-light filters shift colour deliberately and will distort the blue–yellow ladder in particular. We clip the ladder to the smallest step your display can actually render, so the test never asks you to distinguish two identical patches — but everything past that point is your equipment and your room, and we cannot see any of it.
Related tools
The highest frequency you can still hear, measured on a stepped ladder with silent catch trials — and bounded hard by your headphones, your volume and your device's sample rate.
Open tool →Reaction Time TestFive trials measuring how fast you respond to a colour change, reported as a median with the share your own display could account for.
Open tool →Sequence Memory TestThe longest sequence of block positions you can reproduce in order, presented at the rate the board-based research used — and reported as a screen score, not a clinical span.
Open tool →