Can You Train Your Reflexes?
What a Reaction-Speed Test Actually Measures

When you take a reaction-speed test, you get a number: "your reaction time was 240 milliseconds." Then a verdict — faster than average, or slower. But what is that number actually measuring?

I went through the primary literature, and quite a lot of what I found runs against what most of us assume. Elite athletes are not faster at reacting. Raw reaction time barely changes even after years of practice. And there is no such thing as a "reflex nerve."

There is no "reflex nerve"

In Japanese, people say hansha shinkei — literally "reflex nerve" — the way English speakers say "quick reflexes." Neither phrase names anything anatomical. Physiology has two separate concepts here: reflexes and reaction time.

A reflex runs along a fixed path from sensor to muscle and involves no thinking and no deciding. Tap below the knee and the leg kicks — that one. The signal never reaches the brain; it turns around inside the spine, which is why it is called a spinal reflex.

The gap between a stimulus arriving and a muscle beginning to move is called latency. Reflex latencies are short: 25–50 ms in the arm, and 40–45 ms when measured in the calf muscle. A millisecond is a thousandth of a second, so that is 0.025–0.05 seconds — faster than a blink.

Reaction time is the time to a voluntary action. It contains three stages: perceiving the stimulus, deciding and selecting a response, and executing the movement. For a button press, 150–300 ms.

This may sound like semantic nitpicking, but it matters later. Reflexes cannot be trained. Voluntary reactions contain a decision — and where there is a decision, there is room for change.

Reaction time is three times added together

A research team in the United States asked 1,469 people aged 18–65 to do one thing: press a button as fast as possible when a mark appears on screen. Dozens of times over.

The average was 231 ms — 213 ms once the delay of the computer and mouse themselves was subtracted.

The interesting part is the breakdown. The researchers separated stimulus detection time from movement initiation time. Detection averaged 131 ms. The rest was getting the body to move.

That hearing is faster is consistent across studies. A 2024 human neuroimaging study reports response onset at about 25 ms in primary auditory cortex and about 48 ms in primary visual cortex — a median difference of roughly 23 ms.

Two years of practice did not change raw reaction time

Can reaction time be trained? One study followed 94 high-school students (26 in the baseball club, 68 not) for two years.

The same group also compared 82 university students (22 baseball, 22 tennis, 38 non-athletes) with 17 professionals. Again, no differences in simple reaction time by sporting experience or skill level. The differences appeared in Go/Nogo reaction time: baseball players faster than the others, higher skill faster still, professionals fastest.

There are reports that repeating the task itself can shorten simple reaction time somewhat — one study trained 16 people over three weeks. But the sample was small and I could not confirm how many milliseconds were gained.

The finding that reaction time grows with the number of choices is known as Hick's law. But with enough practice the difference between two and four choices disappears — reported in 1959. A 2018 review confirms that practice flattens the slope: in the first session eight choices were about 500 ms slower than two, by the fifth session just over 300 ms.

What shrinks is the deciding. "React faster" is hard to train. "Decide sooner" is not.

Elite athletes are not faster — they can wait

So what is different about elite performers?

A classic cricket study found that professional batsmen's reaction times while batting were about 200 ms — the same as laboratory values. And professionals were no quicker than casual players at picking up information from the bowler. The author concluded that the difference lies not in the speed of the perceptual system but in how the motor system is organised around it.

The football goalkeeper studies are more striking. Watching penalty-kick footage, expert keepers, compared with novices:

• predicted direction more accurately
• waited longer before initiating a response
• made fewer corrective movements
• used fewer, longer fixations, concentrated on the kicking leg, standing leg and ball

Novices spent longer looking at the trunk, arms and hips.

A study of world-class batsmen found that experts were not processing the same cues faster — they had acquired the ability to use earlier cues that less-skilled players were not attuned to (the bowler's hand and arm). Within the same window of time, they take their information sooner.

A meta-analysis in combat sports found the expert advantage larger in response accuracy (effect size 1.24) and fewer fixations (−2.04) than in response time (−1.00). The authors state plainly that expert speed comes from perceptual anticipation using advance cues, not from raw processing speed.

What about FPS and esports?

Surely here reaction time is king. A meta-analysis pooled 15 studies and 1,085 players, comparing professionals and the top 1% against ordinary players:

A separate study of 131 players found no significant difference in choice reaction time between gamers and non-gamers. Comparing 14 professional FPS players with 16 amateurs, processing speed showed no significant difference either; the large gap was in spatial short-term memory (effect size 1.27).

What separates them is not speed of reaction but how they hold space: remembering what was where, tracking several things at once, reading where something will go next. The "good vision of the field" we praise in football and basketball is probably the same family of skill.

Does reaction-speed training work?

In a study where 11,430 people trained online for six weeks, the trained tasks improved but nothing transferred — not even to closely related tasks.

A review of 43 studies on commercial cognitive-training devices found that only one directly tested transfer to sporting performance. A review of one well-known product found just three far-transfer studies, two showing no effect, and not a single preregistered study.

The most useful finding comes from a meta-analysis of 33 randomised controlled trials (RCTs — where participants are split into groups by lot, the most trusted study design) covering 1,048 participants. It split the results by how closely the thing people practised resembled the thing they were tested on:

Their conclusion: the observed improvements "may reflect task familiarity rather than true cognitive enhancement."

That is not the whole story. Training that uses footage from the sport itself — occluding an opponent's action partway and asking what comes next — shows large effects across 12 studies, and the effect survives in field tests, not just video tests. A Japanese study found that video-based perceptual training brought novice batters' prediction accuracy up to expert level.

What works is domain-specific anticipation. What works poorly is generic training disconnected from the domain.

Do we slow down with age?

A UK study of 7,130 people found that simple reaction time shows little slowing until around 50, whereas choice reaction time slows throughout adulthood. Again the split between raw reaction and reaction-with-a-decision.

The magnitude is smaller than people assume: 0.45–0.55 ms per year, about 5 ms per decade. And the slowing is in motor output, not in detecting the stimulus.

Is there a setting where measuring reaction time genuinely matters? One: predicting falls.

In 477 older adults, choice stepping reaction time was a stronger predictor of falls than strength, processing speed or balance. Fallers averaged 1,322 ms, non-fallers 1,168 ms.

For a phrase in such wide everyday use, "reflexes" turns out to be almost absent from official settings. Two things I found interesting in the Japanese sources:

• Japan's national fitness test contains no reaction-time item — in none of its four age bands from 6 to 79. The side-step test measures agility, not response to a stimulus
• The mandatory course for older drivers does not measure reaction time either. It tests dynamic visual acuity, night vision and visual field. The cognitive screening covers cued recall and temporal orientation

So is the number meaningless?

I have spent most of this article on the negative. But I do think there is one thing a reaction test measures well: how the signal is reaching you.

As above, sound and light do not arrive at the same time — about 25 ms versus about 48 ms to sensory cortex. A sound cue and a light cue are not the same cue.

In a classroom or on a sports field this matters more than it looks. "Ready, go" called out loud, a whistle, a clap — when the signal is delivered only as sound, anyone who does not receive that sound is late by definition. And in the record, that lateness is written down as a slow reaction.

The reaction was not slow. The signal did not arrive. Those are different things that produce the same number.

That said — even after all of the above, you probably still want to know your own number. So do I. It feels good when it is fast and annoying when it is not, and that is a perfectly reasonable thing to feel.

So I am not going to say measuring is pointless. The point is to measure in a way that is worth something.

Which is why we changed how we measure

With all of the above in hand, we reworked our reaction checker.

The verdict uses the median — but the fastest attempt is shown too

Out of five attempts, anyone can badly miss one — a slipped finger, a moment of distraction. A mean lets that single attempt drag everything, so the comparison against age-band reference values uses the median.

But wanting to know your single fastest attempt is a thoroughly human thing to want. So the fastest is displayed right alongside it, and personal bests are recorded on that number. There is no reason to hide it. Only the basis for the verdict changed.

We show consistency as well as speed

The five results are plotted as dots with a vertical line at the median, along with the gap between the fastest and slowest attempt.

Speed is partly something you are born with. A tight spread, on the other hand, means you pressed consistently — and that can improve for anyone. With only one axis to be praised on, the game belongs to the fast. So we made it two.

A "colour only" mode to pair with "sound only"

Sound-only (measured with the eyes closed) and colour-only (no sound, no vibration) were designed as a pair from the start. With only one of them, either a deaf child or a child with low vision is shut out of the activity entirely.

Using colour alone would then exclude people who cannot easily distinguish colours. So the cue carries both: a yellow ▲ means "not yet", a green ● means "press now". If the colours are hard to tell apart, the shapes remain.

Trying both modes tells you whether your ears or your eyes are quicker. For most people hearing is slightly quicker, but not for everyone.

We fixed the measurement itself

The tool used to detect a tap on touchend — the moment the finger lifts. That added the entire press-and-release interval, showing results 30–80 ms slower than reality. It now takes the moment of contact, and times the cue from the frame actually painted to the screen. Audio cues are corrected for output latency.

Summary

If someone has told you that you react slowly — or if you have said that to someone — it may be worth asking once what that number was made of. Changing how the signal is given can change the number.

This is one way of looking at it, and the research has limits. Reaction-time measurements shift with stimulus intensity and equipment, and several of the studies cited here have modest sample sizes. If something here does not match what you see in practice, I would be glad to hear it through the contact form.

Sources