Data reference · Benchmarks and measurement

Average Reaction Time

Published benchmarks, age-group data, downloadable tables and measurement notes.

This average reaction time reference brings together published visual-response benchmarks, age-group summaries and the measurement conditions needed to compare average reaction time data fairly.

Recorded mean: 231 ms
Study-corrected mean: 213 ms
Participants: 1,469
Participant ages: 18–65 years

Woods et al. (2015), experiment 1; means rounded to whole milliseconds. The correction applies to the study's equipment. These are published research values, not AimTests visitor statistics. Source [1]

Average reaction time research benchmarks

The figures below concern simple visual responses. Other tasks need their own comparison data; see the definitions and task types for that distinction.

Woods and colleagues measured simple visual responses in two experiments. Participants used a gaming mouse and completed 20 practice trials before the measured trials. The researchers measured hardware delays and reported both recorded and corrected averages. [1]

Published average reaction time benchmarks
Study sample	Measured trials per person	Reported mean
Experiment 1 1,469 adults, ages 18–65	120 simple visual trials	231 ms recorded 213 ms hardware-corrected
Experiment 2 189 adults, ages 18–82	100 simple visual trials	238 ms recorded 220 ms hardware-corrected

Rounded means from two experiments in the same 2015 paper. The second experiment used another sample and a wider range of stimulus intervals. These are research measurements, not averages collected from AimTests visitors.

The corrected figures apply to the researchers' equipment. Subtracting about 18 ms from your browser result would assume your setup has the same delay, which has not been established. Use the figures as research context while keeping your personal comparisons on a consistent setup.

Average reaction time by age

The first experiment also reports seven age groups. Mean responses were slower in the older groups, with an overall age association of about 0.55 ms per year. That is a relationship observed in a sample, not a prediction of how much an individual's result will change each year.

The table below identifies each group by its mean age. Group means and between-person standard deviations are transcribed from Table 2 of the paper, before hardware correction.

Average reaction time by age group
Group mean age (years)	Participants	Mean response	Between-person SD
20.8	86	217.9 ms	19.5 ms
28.6	115	221.0 ms	22.8 ms
35.5	201	224.8 ms	23.4 ms
42.3	273	227.7 ms	26.6 ms
48.6	276	233.6 ms	27.2 ms
55.0	272	236.4 ms	27.0 ms
61.6	246	239.1 ms	28.1 ms

SD means standard deviation. Here it describes differences between participants' mean scores within a group; it is not a device-error allowance or a range that every person must fall within. The groups together cover ages 18–65.

People within the same group had different results. Use these group averages to understand the study's pattern, without turning a group mean into a pass/fail target. These adult data do not supply norms for children, and a single score does not determine a game rank.

Measurement error budget: what can change the reading?

The displayed interval combines your response with the equipment and software used to record it. Some effects create a repeatable offset; others vary between attempts. Research on timing software distinguishes accuracy from precision: consistently similar readings can still include an offset. [2]

The table separates cycle calculations, unknown device delays and effects on which readings are accepted or summarized. The values are not measured extra latency for AimTests, and the rows cannot be added into a personal correction.

Source	Timing scale or effect	Can it be removed?
Display refresh phase	60 Hz: 16.7 ms per cycle; idealized wait averages 8.3 ms within a 0–16.7 ms range. At 144 Hz: 6.9 ms per cycle; idealized mean 3.5 ms.	A higher rate shortens a refresh cycle. It does not establish total display delay or justify subtracting the half-cycle value.
Mouse report phase	125 Hz: 8 ms per report cycle; idealized mean wait 4 ms. At 1,000 Hz: 1 ms per cycle; idealized mean 0.5 ms.	These figures cover only the assumed report phase. Sensor, switch, firmware and connection delays remain unmeasured.
Browser rendering and display processing	No fixed value for an unmeasured setup. A frame callback runs before repaint, not at the moment light reaches your eyes.	A precise browser clock cannot remove this delay. Independent hardware measurements are needed to quantify it.
Anticipation	A guessed response can make a reading unusually short. There is no established fixed correction such as subtracting or adding 50–100 ms.	Random waits and invalid-round rules reduce some predictable or early inputs; they cannot prove every accepted response followed the cue.
Variation, fatigue and repeated readings	Individual readings vary. A five-reading median may differ from a mean or a single result by any amount, including zero.	A median reduces the influence of an isolated slow reading. It does not remove a shared device offset or guarantee a 10–20 ms reduction.

Calculation assumptions: a cycle lasts 1,000 divided by the rate in Hz. If an event arrives uniformly between updates and waits until the next update, its mean wait is half a cycle. Real display scheduling and input reporting need not satisfy that assumption. The refresh and rendering rows can describe overlapping parts of the same process.

For example, the difference between the illustrative 60 Hz and 144 Hz half-cycle values is about 4.9 ms. That is a difference between two model calculations, not a promised improvement from changing your monitor. Actual scanout, pixel response and processing still matter.

How AimTests handles timing and invalid rounds

These are the current rules of the Reaction Time Test. They explain how a reading is produced and filtered. They are site measurement rules, not population norms or a calibration of your device.

Rule	What the tool does	What that means for a score
Random wait: 1.5–6 seconds	Varies the time before the visual cue on each round.	The waiting period is not included in the displayed response interval.
Frame callback and input event	Starts its clock in a frame callback around the visual change and stops on an accepted input event.	The endpoints are browser events, not sensors detecting physical light or the start of finger movement.
Early response and under-80-ms filter	Rejects inputs before the cue and readings below 80 ms.	The cutoff is a filtering rule of this tool. It does not establish that every response below 80 ms is physiologically impossible.
100 ms input deduplication	Ignores inputs within 100 ms of the previous processed input.	This is an input filter, not a 100 ms addition to an accepted reading. It does not identify all automated input.
Hidden tab	Voids an active round when the page becomes hidden.	A round interrupted by switching tabs is excluded rather than paused and resumed.

MDN documents that requestAnimationFrame callbacks run before repaint and are paused in most hidden tabs. [3] This explains why a frame callback is useful for scheduling a visual cue but is not proof of its physical onset time. The rules above describe the reviewed implementation; they do not certify timing accuracy.

Mean, median and personal best

Before comparing scores, check which statistic is being reported. Consider five illustrative readings: 218, 225, 229, 232 and 346 ms.

Mean: 250 ms. Add the five readings and divide by five. Every valid reading contributes.
Median: 229 ms. Sort the readings and take the middle one; for an even count, average the two middle values.
Best: 218 ms. The quickest accepted reading is a personal milestone, not a summary of the whole set.

The slow reading changes the mean more than the median. Choosing the median does not remove hardware delay or guarantee a lower score on the next attempt. Keep slow but valid readings, and compare the same statistic over a fixed number of trials. The example above demonstrates the arithmetic; it is not a population benchmark.

Is your score above or below average?

A browser reading of 250 ms is numerically higher than the first study's 231 ms mean. That comparison alone cannot tell you how you rank: the task and equipment differ, and a mean does not describe a full population distribution. Likewise, a 200 ms reading does not establish an elite gaming level.

For a repeatable way to collect your own readings, use the practice guide. The research tables here explain the comparison; the main test records your individual attempts.

Ready to compare your own readings?

Measure your response

Download and cite the reference data

Use the two CSV files for the published numeric tables, or the JSON file for both tables with field definitions, source details and limitations. Columns identify units and rounding; each CSV row includes the original source DOI and URL.

The downloads contain aggregate research values transcribed by AimTests. They do not contain participant-level records, visitor results, modeled cycle delays or site filtering thresholds. Formatting and field labels were adapted; the experiment means are rounded as indicated. Cite the original study when reusing the values, and this page for the transcription and measurement notes.

The transcription, field labels and file formatting in these downloads are released under the Creative Commons Attribution 4.0 International licence. The figures themselves remain part of the original open-access article, which is distributed under the Creative Commons Attribution licence; credit Woods et al. (2015) when reusing them.

Reference version 1.0 · Reviewed September 19, 2026.

Suggested reference: AimTests. Average Reaction Time: Age Data, Benchmarks & Device Effects. Reference tables transcribed from Woods et al. (2015). https://aimtests.com/average-reaction-time.

Original sources and attribution

Woods DL, Wyma JM, Yund EW, Herron TJ and Reed B (2015). Factors influencing the latency of simple reaction time. Frontiers in Human Neuroscience, 9:131. doi:10.3389/fnhum.2015.00131. Source for both experimental averages, the trial counts and the age-group data. The article is distributed under the Creative Commons Attribution licence.
Bridges D, Pitiot A, MacAskill MR and Peirce JW (2020). The timing mega-study: comparing a range of experiment generators, both lab-based and online. PeerJ, 8:e9414. doi:10.7717/peerj.9414. Source for timing-validation methods and the distinction between offset and variability; it did not test AimTests.
MDN Web Docs. Window: requestAnimationFrame() method. Browser documentation for callback timing relative to repaint.

Sources and AimTests timing rules checked September 19, 2026. The tables identify published research, calculations under stated assumptions and site rules separately. Cycle calculations and the five-reading example are explanations, not new experimental data.

Frequently asked

What is the average reaction time for humans?

For a simple visual task, Woods et al. (2015) reported a mean of 231 ms in 1,469 adults aged 18–65, or 213 ms after correction for that study's hardware delays. This is a published benchmark for a specific task and sample, not a universal average for every person or device.

Is 250 ms a good average reaction time?

A 250 ms browser result cannot be assigned a population rank from one research average. The task, display, input device and reported statistic all affect the comparison. A published mean is context, not a pass/fail line for an uncalibrated browser score.

How does average reaction time change with age?

In Woods et al.'s first experiment, group means increased from 217.9 ms in the youngest group to 239.1 ms in the oldest. The groups covered adults aged 18–65. These are group averages before hardware correction, not targets for each birthday or standards for children.

Should I compare the mean, median or personal best?

Average usually means the arithmetic mean: add the valid times and divide by the number of attempts. The median is the middle sorted time and is less affected by an isolated very slow attempt. Best is the lowest time. Compare the same statistic over the same number of attempts; taking a median does not remove device delay.

Can I subtract monitor or mouse latency from my result?

Not from refresh rate or polling rate alone. The half-cycle examples assume an event is equally likely to arrive anywhere between periodic updates. They are not measurements of your equipment. A valid correction needs independent timing measurements of the complete setup.

Does the 100 ms input filter add 100 ms to my score?

No. It ignores inputs that arrive too close to the previous processed input. It does not hold an accepted response for 100 ms before timing it. Ignoring an input can affect which attempt is recorded, but it is not a constant latency to subtract.

Does taking five readings remove device delay?

No. A median summarizes the middle reading and is less affected by an isolated slow response than a mean. It does not cancel a delay shared by all five readings, and there is no fixed number of milliseconds it will remove.

Why do my results differ across websites or devices?

Sites can use different cues, random waits, input events and rules for accepting attempts. A phone and a mouse also have different display and input paths. Compare results under matching conditions instead of treating all displayed milliseconds as the same measurement.