The Color Discrimination Test, Explained

After reading this you will understand what a color discrimination threshold is, how an adaptive staircase finds yours in about a minute, and how to read your best level without fooling yourself.

What the test measures

Your eye does not see color as a fixed set of labels. It sees differences. Two tiles that look identical at a glance can differ by a small amount of hue and lightness that you either can or cannot detect. The smallest difference you can reliably spot is your discrimination threshold, and this test estimates it.

Each round draws a grid of tiles that share one base color. Exactly one tile is nudged: its hue is rotated by a few degrees and its lightness shifted by a few percent. Your job is to click that odd tile. If you find it, the nudge shrinks and the grid grows. If you miss, the nudge grows back. Over 20 or 30 rounds the difficulty settles near the point where you are right about three quarters of the time.

Here is the hook. Suppose the base tile sits at hue 210 degrees and lightness 55%. Early on the odd tile might be at hue 222 and lightness 61%, a gap almost anyone sees in a 3-by-3 grid. By level 12 the gap can fall to hue 212 and lightness 56% in a 6-by-6 grid. That last difference is close to the limit of human vision on a good screen, and finding it is a real achievement.

When to use it, and when not

Use the timed 60-second mode when you want a single comparable number: how deep into the staircase you can get before the clock stops. Use the untimed practice mode when you want to explore your threshold without pressure, since removing the clock lets you study each grid until you are sure.

This is a screening-style demonstration of visual acuity for color, not a clinical diagnosis. It will not tell you which type of color vision deficiency you have. Formal tests such as the Ishihara plates or the Farnsworth-Munsell 100 Hue arrangement are designed and validated for that, under controlled lighting on calibrated equipment. Treat your score here as a fun personal benchmark and a way to compare screens, not as medical information.

If you notice a large and stable gap between the two halves of your visual field, or you consistently confuse colors that others describe as obvious, see an optometrist. A browser game running on an uncalibrated screen cannot rule anything in or out.

The staircase, and why it beats fixed levels

A naive test would show ten fixed difficulties and count how many you pass. That wastes most of your trials on rounds far too easy or far too hard, which tell you nothing. An adaptive staircase spends nearly every trial near your threshold instead.

The simplest version is the one-up one-down rule: harder after each correct answer, easier after each miss. That rule converges on the level where you are right 50% of the time. This test uses a variant closer to a one-down two-up style, where the difference shrinks after one correct pick but grows more sharply after a miss, which pulls the convergence point up toward the 71% correct level. The general result comes from the transformed staircase rule:

P_{\text{target}} = \left(\tfrac{1}{2}\right)^{1/n}

Here n is the number of consecutive correct answers required before the task gets harder, and P_{\text{target}} is the accuracy the staircase settles on. With n = 1 you get P_{\text{target}} = 0.5. With n = 2 you get P_{\text{target}} \approx 0.707, which is why two-in-a-row rules target roughly 71% correct.

The color difference itself is a single number, a delta, that scales both the hue rotation and the lightness shift. Call the delta at round i the value d_i. A common update multiplies the delta by a factor:

d_{i+1} = \begin{cases} d_i \cdot s & \text{if correct} \\ d_i \cdot g & \text{if wrong} \end{cases}

where s \lt 1 shrinks the delta after a correct pick (say s = 0.85) and g \gt 1 grows it after a miss (say g = 1.3). Because g \gt 1/s, one miss undoes more than one success, so the staircase parks itself above the 50% point.

Worked example: the demo defaults

Run the demo with the default settings. Assume the delta starts at 12 (in the tool's internal difficulty units), the shrink factor is s = 0.85, and the grow factor is g = 1.3. Watch how ten rounds unfold for a player who misses once.

Round 1: delta 12.00, correct. Next delta 12 \times 0.85 = 10.20.
Round 2: delta 10.20, correct. Next 10.20 \times 0.85 = 8.67.
Round 3: delta 8.67, correct. Next 8.67 \times 0.85 = 7.37.
Round 4: delta 7.37, correct. Next 7.37 \times 0.85 = 6.26.
Round 5: delta 6.26, miss. Next 6.26 \times 1.3 = 8.14.
Round 6: delta 8.14, correct. Next 8.14 \times 0.85 = 6.92.
Round 7: delta 6.92, correct. Next 6.92 \times 0.85 = 5.88.
Round 8: delta 5.88, correct. Next 5.88 \times 0.85 = 5.00.
Round 9: delta 5.00, correct. Next 5.00 \times 0.85 = 4.25.
Round 10: delta 4.25, correct. Next 4.25 \times 0.85 = 3.61.

Notice the miss at round 5 did not send you back to the start. It bumped the delta from 6.26 to 8.14, one step easier, and then you resumed shrinking. After ten rounds the delta sits near 3.6, less than a third of where it began. That is the staircase homing in.

The delta falls fast at first, jumps up at the miss on round 5, then resumes its decline. The vertical marker sits on the miss.

Try the staircase yourself

The single idea a static chart cannot show is how the target accuracy depends on the up and down rule. Move the required run length and watch where the staircase settles.

Reading and interpreting your result

Your headline number is the best level you reached, which corresponds to the smallest delta you resolved. Because the staircase reverses direction each time you cross threshold, the honest estimate of your threshold is not the single deepest delta you hit. It is the average of the delta values at the reversal points, ignoring the first two or three reversals while the staircase is still swinging in.

Consider a run with reversal deltas of 6.3, 4.9, 5.5, 4.6, 5.1, and 4.8. Drop the first two as warm-up. Average the last four: (5.5 + 4.6 + 5.1 + 4.8)/4 = 5.0. That 5.0 is a steadier threshold estimate than the lucky 4.6 low point.

Rough interpretation bands for the stored best level
Best level reached	Approximate delta	What it suggests
1 to 5	12 to 6	Warming up, or a low-contrast screen
6 to 10	6 to 3	Typical range on a good display
11 to 15	3 to 1.5	Fine discrimination, near the edge
16+	under 1.5	Exceptional, or an unusually good panel

Your best level is stored on the device you played on. That is deliberate. A comparison is only fair when the screen, brightness and viewing angle are held constant, so the number means something only against your own past runs on the same panel.

Common mistakes that inflate or deflate your score

The biggest source of error is the screen, not your eyes. On a cheap TN laptop panel, tilting the lid five degrees can shift the visible lightness of a tile by more than the delta you are trying to detect. That makes deep levels either impossible or, worse, guessable from a viewing-angle artifact rather than real color perception. An IPS or OLED panel at a fixed angle gives a fair test.

Screen brightness matters just as much. At 20% brightness in a dim room the tiles crush together and small deltas vanish. At 100% in bright sun, glare washes out the same deltas. Pick one comfortable brightness and keep it across sessions.

Two behavioral traps remain. First, guessing when unsure. A random click in a 6-by-6 grid is correct 1 in 36 times, so lucky guesses barely help, but they push the staircase harder than your true threshold and cost you on the next round. Second, rushing in timed mode. Speed helps your level count only if accuracy holds; a single careless miss triggers the grow factor and undoes several rounds of progress.

A typical psychometric curve. Below a delta of about 2 you are near chance; above about 6 you almost never miss. The steep middle, marked near delta 3.5, is where the staircase spends its time.

Related tests on this site

Color discrimination is one slice of sensory performance. The closest relative here is the Contrast Sensitivity Test, which uses the same staircase logic to find the faintest pattern you can see rather than the smallest color difference. For hearing, the Hearing Frequency Test maps the highest pitch you can detect across octave bands.

If you enjoy timed self-benchmarks, the Reaction Time Test measures raw response speed in milliseconds, and the N-Back Working Memory Test probes short-term memory instead of the senses. All of them share the idea that a fair number needs consistent conditions from run to run.

Frequently asked questions

Does this test diagnose color blindness?

No. It measures how fine a difference you can resolve on your screen, not which cone types you have. Color vision deficiencies affect specific hue directions in ways this single-delta test does not isolate. For a real assessment, see an eye-care professional who uses validated plates and arrangements under controlled lighting.

Why is my score so different on my phone versus my laptop?

Panels differ enormously. Contrast ratio, color profile, peak brightness and viewing angle all change how visible a given delta is. A high-quality OLED phone can beat a budget laptop by several levels for the same eyes. Compare only runs made on the same display at the same brightness.

Why did one miss cost me several rounds?

The grow factor after a miss is larger than the shrink factor after a success. That is intentional. It keeps the staircase parked near your 71% accuracy point instead of drifting down to a level where you are only guessing right half the time.

Should I use timed or untimed mode?

Use timed mode for a single comparable score and a bit of pressure. Use untimed practice mode to study each grid and find where your threshold really sits, since removing the clock lets you take a careful look before every click.

What is the smallest color difference a person can see?

Under ideal conditions the just-noticeable difference is a tiny fraction of the early-round nudges in this test. Reaching the deepest levels means resolving hue and lightness shifts near that biological limit, which is why late levels feel almost impossible and why screen quality decides whether they are reachable at all.

Try it yourself in the tool