Direct answer
A reaction time percentile tells you what fraction of people you are faster than. If your score sits at the 90th percentile (p90), you are faster than 90% of test-takers; at the 50th percentile (p50, the median) you are squarely average; at the 10th percentile (p10) you are faster than only 10%, meaning most people beat you.
Reaction time is reported in milliseconds, and lower is faster, so the percentile axis runs opposite to the time axis: low millisecond values map to high percentiles. On the large public dataset behind Human Benchmark (over 81 million simple-reaction tests), the median is about 273 ms and the mean is about 284 ms. The distribution is right-skewed, which is why those two numbers differ and why the median, not the mean, is the honest center of the pack.
The hard limit at the fast end is physiology. Genuine simple reaction times bottom out around 100-120 ms; anything materially below that is anticipation (you moved before you truly processed the signal), not reaction. That floor is why a percentile near the top compresses sharply: the population cannot get much faster, so small millisecond gains buy large rank jumps.
What each percentile means
A percentile is a rank, not a score. It answers "how do I compare?" rather than "how fast am I in absolute terms." The two questions are related but distinct: your millisecond number is fixed, but your percentile depends entirely on the population you are compared against.
Read the table below as a position on the curve. The millisecond bands are illustrative of the shape of a typical browser simple-reaction distribution (visual stimulus, mouse or tap input), anchored to the publicly reported ~273 ms median. They are not a verdict on whether a score is good for your age — that depends on cohort, which is covered separately.
| Percentile | Meaning | Where it sits on the curve |
|---|---|---|
| p10 | Faster than 10% of people | Slow tail — most people beat you |
| p25 | Faster than 25% | Below median, lower quartile |
| p50 (median) | Faster than half | The center of the pack (~273 ms) |
| p75 | Faster than 75% | Upper quartile, clearly above average |
| p90 | Faster than 90% | Fast tail — top 10% |
Why the distribution is skewed (and why median beats mean)
Reaction times do not form a symmetric bell curve. They have a sharp rise on the left and a long tail to the right — a shape statisticians model with the ex-Gaussian distribution. There is a hard biological wall stopping anyone from being absurdly fast, but no wall stopping a trial from being slow: a lapse in attention, a tired moment, a laggy device, or a fumbled click all push individual times far to the right.
That asymmetry has a practical consequence. The mean gets dragged toward the slow tail, so it overstates the typical experience — which is exactly why the public 273 ms median sits below the 284 ms mean. For skewed data, the median is the more faithful summary of where most observations actually land, a point made directly in the reaction-time methods literature. When you read a percentile chart, you are reading position relative to that median-centered curve, not relative to an inflated average.
It also explains why percentiles are spaced unevenly in milliseconds. Near the median, a large mass of people is packed into a narrow band, so a 20 ms improvement can move you several percentiles. Out in the fast tail, the population thins toward the floor, so each additional millisecond of rank costs progressively more.
The physiological floor: ~100-120 ms
Every reaction-time percentile curve runs into the same wall on the left. Signal has to travel from sense organ to brain, get processed, and return as a motor command. For a simple reaction that round trip cannot realistically finish faster than roughly 100-120 ms.
Elite sport makes the floor concrete. World Athletics treats any sprint start under 100 ms after the gun as a false start, on the reasoning that a sub-100 ms response reflects anticipation of the gun rather than a genuine reaction to it. Research on elite sprinters has questioned whether the exact cutoff should be lower, but the principle holds: below the floor you are predicting, not reacting.
This matters for reading your own percentile. If a browser test ever shows you a number near or below ~120 ms on a simple-reaction task, the likely explanation is a pre-emptive click rather than a record-breaking nervous system. Honest fast scores cluster against the floor without piercing it, which is why the top percentiles bunch tightly together.
Reading your own percentile without fooling yourself
Two caveats keep a percentile honest. First, modality and hardware shift the whole curve. People react faster to sound than to light — auditory simple reaction times tend to run a few tens of milliseconds quicker than visual ones — and a browser test adds display and input latency on top, often on the order of ~30 ms. So a visual browser percentile is not directly comparable to a lab auditory measurement; compare like with like.
Second, one trial is not your percentile. Because the right tail is fat with lapses, a single slow attempt can drop your apparent rank sharply. Take several trials and read your median against the distribution, not your worst run and not your single lucky best. For sustained vigilance over many trials — closer to how attention behaves in the real world — a longer protocol like the psychomotor vigilance task is the better lens.
Deadline first-party data (coming soon) — percentile bands (p10/p25/p50/p75/p90) computed from N real runs in our database, so you can read your rank against actual Deadline test-takers rather than a borrowed dataset.
Frequently asked questions
What percentile is a good reaction time? As a rank, anything above the 50th percentile means you are faster than most people, and the 90th percentile (top 10%) is genuinely fast. But "good" depends on your cohort — age in particular shifts the whole curve. Percentile answers "how do I compare," while "good" is a separate, cohort-dependent question.
What does the 50th percentile (median) reaction time mean? The median is the exact middle of the distribution: half of people are faster and half slower. On the public Human Benchmark dataset of 81M+ simple-reaction tests, that median is about 273 ms. Because the distribution is right-skewed, the median is a more honest center than the mean (~284 ms), which the slow tail pulls upward.
Why are reaction time percentiles spaced unevenly in milliseconds? The distribution is dense near the median and thins toward the fast end as it approaches the physiological floor. Near the middle, a small millisecond improvement passes many people and moves you several percentiles. In the fast tail, the population is sparse, so each percentile costs more milliseconds.
What is the fastest a human reaction time can be? Genuine simple reaction times floor out around 100-120 ms. World Athletics flags any sprint start under 100 ms as a false start because such a response reflects anticipation rather than a true reaction to the stimulus. Browser scores near or below ~120 ms usually indicate a pre-emptive click, not record speed.
Why does my browser reaction time percentile differ from a lab test? Browser tests measure visual reaction plus display and input latency (often ~30 ms), and visual reactions are inherently slower than auditory ones by a few tens of milliseconds. A visual browser percentile and a lab auditory measurement sit on different curves, so compare like with like.
Related on Deadline
Reaction Time Test · Psychomotor Vigilance Test · average reaction time by age · what counts as a good reaction time · reaction speed tests
Sources and notes
- https://humanbenchmark.com/tests/reactiontime/statistics
- https://www.biorxiv.org/content/10.1101/383935.full.pdf
- https://garstats.wordpress.com/2018/02/02/rtbias1/
- https://www.basvanhooren.com/is-it-possible-to-react-faster-than-100-ms-in-a-sprint-start/
- https://pubmed.ncbi.nlm.nih.gov/17127583/
- https://www.scirp.org/html/4-2400003_2689.htm
- https://pmc.ncbi.nlm.nih.gov/articles/PMC4988978/
- https://link.springer.com/article/10.3758/s13428-020-01466-5