A typing test does not count words. It counts keystrokes, space bar included, and divides by five, so a "word" in a words-per-minute score is any five keys. We counted keystrokes in two nineteenth-century novels and two internet standards, and none averages five a word once the space is counted. Dickens's A Tale of Two Cities came lowest, at 5.55 keystrokes a space-separated word. RFC 9110, the specification for HTTP, came highest at 6.64. On those texts a wpm figure counts a ninth to a third more "words" than a typist produced. The convention is working as written: five fits an English word without its space.
The count is not new. Revising Ida McLenan Cutler's Rational Typewriting for Gregg (copyright 1916 and 1917), Rupert P. SoRelle wrote on page 99 that words in ordinary matter "will average about 5.5 strokes each, counting the spaces". Our two novels land within 0.15 of that.
Typewriter strokes grouped in fives defined the contest word
The earliest explicit rule we found is in Mrs Smith Clough's A Typewriting Catechism (5th edition, Pitman, London, 1930). Its page 131 reports the International Speed Contest Rules: gross strokes were "divided by five to secure the number of gross words", with deductions for errors taken from that total. As the book is a fifth edition, the rule was in print by 1930; we found nothing naming who chose five, or when.
I. Scott MacKenzie of York University defines a word as five characters, spaces and punctuation included, a practice he dates to early typing competitions; our guide to Chinese typing speed applies it to Chinese. Intersteno's 2026 rules for its internet keyboarding contest score in characters a minute instead, and now count one mistake per word, where before computers they counted one per six characters.
Spaces bring real English closer to six keystrokes a word
We stripped licences and page headers from the Project Gutenberg texts and the RFCs, collapsed whitespace, and divided characters plus one space per word by the word count.
| Text | Words | Letters a word | Keystrokes a word, space included | With Shift added |
|---|---|---|---|---|
| Pride and Prejudice (1813) | 127,360 | 4.39 | 5.64 | 5.81 |
| A Tale of Two Cities | 135,886 | 4.22 | 5.55 | 5.72 |
| RFC 8259 (JSON) | 3,843 | 5.00 | 5.92 | 6.43 |
| RFC 9110 (HTTP) | 68,887 | 5.14 | 6.64 | 7.07 |
Without the space, the novels run 4.55 to 4.64 characters a word, and Peter Norvig's count of Google Books words gives "4.79 letters per word". The rule, though, counts the space.
The 10FastFingers FAQ counts Shift, so "America" needs 8 keystrokes, and Intersteno gives each character 1 to 4 points by the keys needed. Our last column adds one key per capital and shifted symbol on a US layout, our assumption. Reading speeds, covered in our guide to reading time in English and Chinese, are a separate measure.
A wrong letter wipes ten words off a typing-contest score
Clough's page reports that "Ten words are deducted from the total" for each of ten faults and that "One error only is counted in any one word." SoRelle gives the same rule and notes that experts abandon a word once they spot a mistake. The penalty outlived the typewriter: a Unix typing script by Steve Hayman of Indiana University, dated 16 August 1988, states "5 chars is a word, 10 WPM penalty for error". Intersteno's 50 points per error equal ten words when each character is worth one point, as lowercase letters are.
We generated a synthetic minute of typing at a true 60 wpm, with the last letter wrong in every tenth word, four uncorrected errors, and applied each rule ourselves, without running any test's software.
| Rule, applied by us | Result for the same minute |
|---|---|
| Speed, errors reported separately (the studies below) | 60 wpm, 1.35% errors |
| Minus one word per error, repeated on calculator pages we checked | about 56 |
| Monkeytype's published definition | about 54 |
| 1930 contest rules, minus ten words per error | about 20 |
| Intersteno 2026, minus 50 points per error | about 97 points a minute, roughly 19 words |
Monkeytype's About page counts only the characters "in the correctly typed words (including spaces), divided by 5", so a wrong word earns nothing. Intersteno's contest runs ten minutes, and 1.35% exceeds its error cap for every category. The same minute comes out anywhere from about 20 to 60.
Backspaced errors vanish from some accuracy figures
Monkeytype defines accuracy as the "percentage of correctly pressed keys", so a deleted key still counts; Palin et al. report 2.3% uncorrected errors, a figure counting only what was left in the text.
Until 19 October 2026 our typing speed test kept only the text left in the box, so a backspaced typo left accuracy at 100%. Net credited correct letters inside wrong words, so typing just the first letter of every word scored 30. It compared letters by position, so one dropped or doubled letter marked the rest of that word wrong, and a perfectly typed quote read 1.1 to 2.5% fast. The page also carried unsourced claims such as a word list covering "85% of everyday English" and a 40-wpm average typist.
The revision made that day subtracts one word a minute for each word left wrong, the minus-one rule above, aligns each word with its target so one slip costs one letter, and counts every key pressed in accuracy. On our synthetic minutes, first letters only now gives net 0 and one typo in ten words gives 55. The test's 1,000-word list averages about 6.2 keystrokes a word, its 40 quotes about 5.2.
The opening keystroke triggers the clock and goes untimed itself
A clock that starts on the first key gives that key no time, and MacKenzie's note says "the character count should be decremented by one". One keystroke in 100 is 1%, and one in 25 is 4%, by our arithmetic.
Dhakal et al.'s CHI 2018 paper times typing "from the first to the last keypress" and reports a mean of 51.56 wpm (SD 20.2) over 168,960 participants, self-selected from a commercial typing site's users, mainly younger people in the US. Palin et al. (2019), drawing on the same company's users, found 36.2 wpm on smartphones among 37,370 volunteers.
Typos beyond a fixed allowance fail a US federal typing test outright
The US Office of Personnel Management's group qualification standard asks Clerk-Typist and Office Automation candidates for "40 words per minute typing speed", with a footnote: "based on a 5 minute sample with three or fewer errors." A fourth error fails the sample whatever the speed; nothing is subtracted, and OPM does not say how it counts a word.
In Hong Kong, the government's youth.gov.hk portal, last updated 12 October 2023, lists an English word processing speed of 20 words per minute for Clerical Assistant candidates, beside a Chinese figure of 15 in a different unit. The Civil Service Bureau's August 2022 press release describes a Chinese and English word processing speed test but no rule for counting a word or an error. We found none elsewhere, and no public Singapore civil-service typing figure.
View author profile → · Editorial policy
