How WPM Is Actually Calculated

The first convention is the one nobody argues about. A word, for the purpose of typing speed, is five characters — and the space that follows it counts as one of them.

So the formula for gross speed is:

gross WPM = (total characters typed ÷ 5) ÷ minutes elapsed

The reason for a fixed length rather than counting actual words is comparability. Real words vary from one letter to fifteen, so counting them would make a passage of short words score far higher than a passage of long ones for identical finger work. Fixing the unit at five characters makes the measure about keystrokes rather than vocabulary, which is what you want if the number is meant to describe typing.

The practical consequence is that the figure is only meaningful for text whose average word length is close to English prose. It is why a numbers test or a code test produces a figure that systematically understates the work done: a line of source code packs far more distinct decisions into five characters than the word 'about' does.

Gross speed counts everything you typed. Net speed subtracts something for errors, and 'something' is where the disagreement lives.

The two conventions in wide use are:

net WPM = gross WPM − (uncorrected errors ÷ minutes)

which charges one word of speed per error per minute, and

net WPM = (correct characters ÷ 5) ÷ minutes

which simply does not count incorrect characters at all. These produce different numbers from the same session, and neither is wrong — the first treats an error as a penalty, the second treats it as work that did not happen.

The first is harsher at speed. Because the penalty is per minute rather than per character, a fast typist and a slow one making the same number of errors in the same minute lose the same absolute amount, which is a larger proportion of the slow typist's total and a smaller one of the fast typist's. The second scales naturally with output. When a site reports a net figure without saying which it used, you cannot reconstruct the calculation, and that alone accounts for a good deal of cross-site disagreement.

Whether an error is charged at all depends on when the site counts.

One approach counts keystrokes: every incorrect key you press is an error, whether or not you fixed it afterwards. Under this rule, a typist who makes twenty mistakes and corrects all twenty has made twenty errors.

The other counts the final text: an error is a character that is still wrong when time runs out. Under this rule, that same typist made zero errors — they paid for the corrections in time, and the time is the only cost.

The gap between those two readings of the same session is often enormous, and it is the single largest source of confusion about typing scores. A typist who corrects everything scores very well under the second and mediocre under the first, with identical typing.

This site counts uncorrected errors — what is still wrong at the end — which means corrections cost you time rather than accuracy. The accuracy mode test is the extreme version of the same idea: you cannot advance past a mistake at all, so error rate converts entirely into elapsed time.

A one-minute test is not necessarily sixty seconds of typing. Some sites start the clock when the page loads, some when you press the first key, and some when you click into the field.

The difference is the time you spend reading the opening words before you begin, which is a few seconds and therefore several percent of a one-minute test. On a fifteen-second test it can be a fifth of the whole measurement.

This site starts on the first keystroke, so reading the opening line costs nothing. That is the more forgiving convention and it is worth knowing when comparing a result here against one taken elsewhere, because a site that starts on page load will report a lower number for the same typing.

The same applies at the other end. A test that stops when the passage runs out measures something slightly different from one that runs the full duration, since a fast typist reaching the end early has their time cut short in a way the arithmetic may or may not handle.

Suppose you type 240 characters in 30 seconds, and 8 of those characters are still wrong at the end.

Gross: 240 ÷ 5 = 48 word-units, over 0.5 minutes, so 96 gross WPM.

Net under the correct-characters convention: 232 correct characters ÷ 5 = 46.4, over 0.5 minutes, so 92.8 — call it 93.

Net under the per-minute penalty convention: 8 errors in half a minute is 16 errors per minute, so 96 − 16 = 80.

Three defensible numbers from one session: 96, 93 and 80. Nothing has been miscalculated. A site reporting any of them is reporting the truth about your typing under a stated convention, and the spread is exactly why comparing a figure from one site to a figure from another is a weaker comparison than most people assume. That argument is developed further in why sites score typing differently.

Words per minute compresses a great deal into one figure, and several things that matter are not recoverable from it.

It cannot distinguish even typing from bursts and stalls. Two typists averaging the same figure can have entirely different distributions, and the burst-and-stall one has a much lower ceiling, which is why the steady rhythm drill exists as separate work.

It cannot tell you which keys cost you. A single figure averages over the whole keyboard, so a specific weak finger disappears into it. That is what per-key error data is for, and why the weak-key diagnostic uses text with even key frequency rather than natural English.

And it says nothing about accuracy, which is reported separately and is itself defined more than one way. The three incompatible definitions in common use are set out in what a typing accuracy percentage means.