The evidence base

Typing advice on the web is mostly folklore repeated until it sounds like fact. This page collects what has actually been measured, with the numbers, so anything we claim can be traced to a source or dropped.

Compiled 12 August 2026 · figures read from the published papers

Mean speed
51.56 WPM
Largest sample
168,960
Touch vs self-taught
No difference
Papers summarised
2

How to use this page. Every figure below is quoted from a peer-reviewed paper, with the DOI given. Where we could not find a study for a common claim, the claim is listed under what is not established rather than quietly repeated. If you find an error in a transcribed figure, that is a bug — tell us.

Contents
  1. Population-scale figures (n = 168,960)
  2. Motion-capture findings (n = 30)
  3. Earlier laboratory studies
  4. What TypeCrt built on this
  5. What is not established
  6. Full citations

Population-scale figures

Dhakal, Feit, Kristensson and Oulasvirta (2018) analysed 136,857,600 keystrokes from 168,960 participants who each transcribed 15 sentences. It is the largest published dataset on typing, and the dataset itself was released for scientific use.

Speed distribution

MeasureValue
Mean words per minute51.56 (SD 20.2)
Fastest 10% of participantsabove approx. 78 WPM
Slowest 10% of participantsbelow approx. 26 WPM
Fastest observed120 WPM and above
Definition of a wordAny five characters

Inter-key interval

The time between consecutive keystrokes is the mechanical substrate of speed, and it correlates with WPM at r = −0.84 — strongly and negatively, as expected.

GroupMean IKIStd. dev.
All participants238.66 ms111.60
Fastest 10%121.70 ms11.96
Slowest 10%481.03 ms123.36

Note the standard deviations: fast typists are not only quicker between keys, they are dramatically more regular — an SD of 11.96 ms against 123.36 ms. This is the empirical basis for treating consistency as a distinct metric rather than a curiosity.

Training

GroupMean WPMStd. dev.
Had taken a typing course (72% of sample)54.3520.80
Had not49.0019.73
Difference 5.35 WPM — Cohen's d = 0.27, a small effect

Other reported findings

Motion-capture findings

Feit, Weir and Oulasvirta (2016) took a much smaller sample — 30 participants, 17 female, aged 20 to 55 with a mean of 31 — but instrumented them properly: optical motion capture on the hands, plus eye-tracking glasses to record where attention went. About 43% self-reported touch typing.

Touch typing versus self-taught

MeasureTouch typistsNon-touch typists
Mean WPM57.83 (SD 15.25)58.93 (SD 10.82)
Mean inter-key interval176.39 ms (SD 44.31)168.91 ms (SD 33.22)
Uncorrected error rate0.76% (SD 0.62)0.47% (SD 0.42)
Weekly typing hours47.1544.56

The WPM difference was not statistically significant (Mann–Whitney U = 103, p = 0.38), and the non-touch group was nominally faster. The paper states that regardless of the number of fingers involved, an everyday typist may achieve rates over 70 WPM, and that some participants using only one or two fingers per hand reached performance “normally attributed to touch typists.”

The three predictors of performance

What did separate fast from slow, per the motion-capture data:

Earlier laboratory studies

Context for why older figures differ. Pre-2000 typing research studied professional typists on typewriters, not the general population, which is one reason the numbers quoted around the web are inconsistent.

These are reported in the related-work sections of the two papers above, which cite the primary sources.

What TypeCrt built on this

We are not claiming the research endorses this product. We are showing which design decisions follow from it, so you can judge whether the reasoning holds.

FindingWhat we did with it
Specific letter pairs predict speed more than general fluency KeyForge scores each key from your own history and weights generated words toward the weakest, instead of drilling everything equally
Fast typists have far lower IKI variance (SD 11.96 vs 123.36 ms) Consistency is reported as a first-class metric next to WPM, not hidden — formula published
A word is defined as five characters in text-entry research The same convention, so a score here is directly comparable to the published percentiles
Faster typists make fewer errors; errors cost time twice Raw WPM is shown alongside net WPM — the gap between them is what your errors cost
Finger count does not predict speed; consistency of mapping does We do not enforce or score “correct” fingering, and do not claim touch typing is required to get fast

What is not established

Claims we see constantly and will not make, because we could not find support for them.

Full citations

Primary sources

Dhakal, V., Feit, A.M., Kristensson, P.O. and Oulasvirta, A. (2018). Observations on Typing from 136 Million Keystrokes. In Proceedings of the 2018 CHI Conference on Human Factors in Computing Systems. ACM. DOI: 10.1145/3173574.3174220. Dataset: userinterfaces.aalto.fi/136Mkeystrokes

Feit, A.M., Weir, D. and Oulasvirta, A. (2016). How We Type: Movement Strategies and Performance in Everyday Typing. In Proceedings of the 2016 CHI Conference on Human Factors in Computing Systems. ACM. DOI: 10.1145/2858036.2858233. Dataset: userinterfaces.aalto.fi/how-we-type

Both papers are from the Aalto University User Interfaces group with the University of Cambridge, and both released their datasets publicly — so every figure on this page can be checked at source rather than taken from us.

See where you fall in that distribution

One 60-second test, scored with the same five-characters-per-word definition the research uses.

Take a typing test →
TypeCrt Docs Blog About Privacy