By Eric, BlinkBench Founder · Last reviewed: September 2026
Take five reaction time readings on yourself right now. The fastest and slowest will probably sit 50 to 100 ms apart. Now compare your median to a friend's median from their own five trials. That gap is almost certainly smaller than the spread inside either of your own runs. This is the central, underappreciated fact about reaction time: the noise within a single person usually swallows the signal between people.
The numbers that make this concrete
Simple reaction time — wait for a stimulus, respond as fast as you can — averaged 231 ms across 1,469 adults aged 18 to 65 in one large study (213 ms once the measured delay of the monitor and mouse was taken out; Woods et al., 2015). Individual trials scatter widely around a person's own average, even with no change in effort or alertness.
The between-person gap is smaller than most people expect. In the same study, people's average reaction times differed from one another with a standard deviation of 27 ms. A single person's trials varied far more: the within-person standard deviation averaged 40 ms in one experiment and 53 ms in a second. The spread inside one person was bigger than the spread between people.
This matters when you sit down to measure your reaction time and immediately want to compare your score to someone else's. If each of you took only one trial, the difference between your numbers reflects your states at that exact moment far more than it reflects any stable difference between you.
Why within-person variance is so large
Reaction time is not a fixed property of a nervous system. It is the output of a process that competes with everything else the brain is doing at that instant.
Alertness fluctuates on a sub-minute scale. Even during a task that lasts only a few minutes, attention drifts. A trial that catches you mid-drift can be tens of milliseconds slower than one that catches you at peak readiness, with no sense of the difference from the inside.
Anticipation and timing interact. When the interval between the warning signal and the stimulus varies randomly, some trials will catch you slightly early in your anticipatory build-up and some slightly late. The preparatory state of your motor cortex at the moment of the stimulus is not identical across trials.
Motor execution adds its own noise. The finger does not always move from the same starting position, with the same muscle tension, or with the same contact geometry on the input device. These physical differences add a few milliseconds of variance that are invisible from the outside.
The display and input chain are part of the measurement. A monitor that refreshes at 60 Hz introduces up to 16.7 ms of frame-timing uncertainty on any given trial, independent of anything you do. A 144 Hz display cuts that to about 7 ms. The device you are using is always part of what is being measured, not a neutral conduit.
What a single comparison actually tells you
Suppose your one-trial score is 210 ms and a friend's is 240 ms. The tempting conclusion is that you are 30 ms faster. But if each of you has a within-person standard deviation around 40 ms, the average Woods and colleagues measured, that 30 ms difference is well within the noise of a single observation for either of you. You cannot distinguish a real stable difference from the ordinary fluctuation that both of you show across trials.
This is not a hypothetical concern. It is the routine situation whenever two people compare single scores. The between-person gap that would be detectable above the within-person noise needs to be large — think elite athlete versus untrained adult, or a clinical population versus controls — before a one-trial comparison carries much information.
What comparing your own trials does tell you
Because the within-person variance is large, it is also informative. Your spread across five trials is a real signal about your consistency, your alertness, and whether something unusual happened on a particular trial.
A median is more stable than a single reading, which is why the Reaction Time Test on this site scores you on the middle trial of five rather than the fastest or the mean. The fastest trial can be a partial anticipation or a lucky coincidence of timing; the mean is pulled upward by the occasional slow outlier. The median is the most honest single-number summary of where you actually sat during that session.
Watching how your own median shifts across sessions — different times of day, different fatigue levels, different days — is where reaction time data becomes genuinely useful. If your median is consistently 30 ms slower in the late afternoon than the morning, that is a real within-person effect. If one session shows an unusually wide spread, something disrupted your consistency that session. These comparisons are meaningful because you are controlling for the between-person variation entirely.
The same logic applies to other speed measurements. The guide on why Fitts'-law throughput survives hardware swaps while raw milliseconds don't makes a related point: a number that stays stable across conditions tells you something about the person; a number that shifts with every equipment change tells you mostly about the equipment.
Frequently asked questions
How many trials do I need before my median is stable?
Five trials is enough to get a median that is reasonably resistant to single outliers, but it is not enough to pin down your true central tendency precisely. With a trial-to-trial standard deviation around 40 ms, the median of five trials has a standard error of roughly 20 ms (about 1.25 × 40 ÷ √5), so running the same session again could easily move it that much. Ten to fifteen trials would tighten that considerably, though most people find that many trials in a single session introduce fatigue effects of their own.
Does faster always mean better?
Not without qualification. A very fast trial is often a near-anticipation rather than a genuine response to the stimulus. The test on this site discards results below a plausible threshold for exactly this reason. A consistent median in a reasonable range is more meaningful than a single fast outlier.
Why does my reaction time feel the same every trial even though the numbers vary?
The differences involved are a few tens of milliseconds from one trial to the next, and nothing about pressing a key tells you how long it took. This is part of why measurement matters: intuition about speed is unreliable at these scales.
Can I compare my score to published averages?
With caution. Published averages come from specific equipment, specific protocols, and specific populations. A figure from a laboratory study using a dedicated response box and a controlled environment is not directly comparable to a score from a consumer monitor and a trackpad. The within-person variability discussion above applies equally to between-study comparisons.
Does reaction time change with age?
Yes. Across adults aged 18 to 65, simple reaction time rose by about half a millisecond per year of age, from 218 ms in the youngest group to 239 ms in the oldest (Woods et al., 2015). Trial-to-trial spread grew only weakly with age in that study. The gap between age groups is real and detectable with enough trials, even though a single comparison between any two individuals remains noisy.
Within-person variance is one of four things separating a score from the person who produced it — see how to read your BlinkBench score for the other three.