What WPM actually means
WPM stands for words per minute: how much text you can produce in a minute of typing. The complication is that “word” is not a usable unit for measurement. Real words vary from one character to fifteen, so counting them would make your score depend mostly on the vocabulary of the passage you were given rather than on how fast you type.
Typing measurement solves this with a fixed-length standard word, and then counts characters instead of words. WPM is therefore a normalised character rate presented in a unit people find intuitive.
The five-character word convention
A common convention in typing measurement is that one word equals five characters, including spaces. Using the same fixed unit is what makes the convention useful: a score from one passage is easier to compare with a score from another.
Every character counts the same under this rule — letters, digits, punctuation and spaces. There is no weighting for difficult keys, because a browser cannot know which keys are difficult on your particular layout and keyboard.
The formula
Testpartout computes:
WPM = (correct characters / 5) / elapsed minutes
Two decisions inside that formula do the real work: which characters are counted, and which elapsed time is used. Both are described below.
Only correct characters count
The numerator is the number of characters you typed that match the target text at their position. Characters that do not match contribute nothing to the score.
The reason is straightforward: if every keystroke counted, mashing random keys quickly could produce a very high WPM and a meaningless result. Typing speed is useful only when it is tied to producing the intended text, so a correct-character basis means the reliable way to raise the number is to type the right characters faster.
Accuracy is reported separately
Because errors are excluded rather than penalised twice, accuracy is shown as its own figure. A high WPM with poor accuracy tells a different story from the same WPM with clean accuracy — the first may mean fast output that needs fixing, while the second is closer to usable text. Reading both together is the point.
Cumulative mistakes
The error count is cumulative: every incorrect character you enter is counted when it happens, and correcting it afterwards does not remove it from the tally. This is deliberate. An error that you noticed and fixed still happened, and it still cost you time. A count that could be erased by backspacing could flatter sessions where mistakes are corrected quickly and hide how often incorrect characters were entered.
Backspace and corrections
You can correct freely. Backspace removes the character from your input, so the text you go on to produce can match the target again, and previously mistyped positions that you retype correctly do count towards the correct-character total.
What correcting does not do is rewind the clock or the error count. The time spent backspacing and retyping is time inside the session, so heavy correcting lowers WPM through time cost rather than through a penalty — which is exactly how it works when you are typing something real.
The 60-second run
The Testpartout typing test is a fixed 60-second session, and the clock starts on your first printable character rather than when the page loads. Nobody should be penalised for reading the prompt, settling their hands, or deciding to start a moment later.
A fixed duration keeps the denominator constant, which makes repeated runs directly comparable and removes the “short passage” effect where a burst of speed over a few words looks like a sustainable rate. Sixty seconds gives a longer sample than a short sprint while still being brief enough for repeated runs.
The deadline is authoritative: the session ends at 60 seconds of elapsed measured time, and elapsed time is clamped to move only forwards, so a suspended or throttled tab cannot produce a shorter-than-real denominator and an inflated score. If the page stops being visible, the run is cancelled rather than scored, because a test you cannot see is not a test of your typing.
A worked example
Suppose that in a 60-second run you enter 260 characters in total, and 240 of them match the target text at their position.
- Correct characters: 240
- Standard words: 240 / 5 = 48
- Elapsed time: 60 seconds = 1.0 minute
- WPM: 48 / 1.0 = 48
The 20 incorrect characters do not add to the score; they appear in the cumulative mistake count and reduce accuracy. Had all 260 characters been correct, the same run would have produced 260 / 5 = 52 WPM. The gap between those two numbers is the cost of errors in this scoring method, and it is why pushing for raw speed past your accuracy limit may not improve the final result.
Physical keyboards versus on-screen keyboards
The same formula applied on a phone measures a different activity. On a physical keyboard you use multiple fingers in parallel with tactile feedback and, usually, no prediction layer between your hands and the text. On a touchscreen you are tapping a flat surface, typically with one or two thumbs, often with autocorrect and word prediction involved.
Neither result is wrong, but they are not the same measurement, and a phone score should not be read as a slower version of a desktop score. Compare like with like: the same keyboard, the same device.
Input-method and browser limitations
- Composition-based input. Input method editors for Japanese, Chinese, Korean and similar systems commit text in blocks rather than one keystroke at a time. The test works from the committed value of the field, which is the only reliable signal available, but character-by-character timing for composed text is inherently coarser.
- Prediction and autocorrect. Where the platform inserts or rewrites text, what arrives in the field is not purely what you typed.
- Paste is blocked. Pasting or dropping text would produce a fictional score, so it is not accepted during a run.
- Keyboard layout and hardware. Layout familiarity, key travel and switch type all affect real typing speed and are not something a web page can normalise away.
- Timing resolution. Browsers deliberately coarsen high-resolution timers for privacy reasons. This is negligible at the scale of a 60-second session, but it is part of why we describe these as practical browser measurements rather than instrument readings.
Comparing your own results
Testpartout publishes no average WPM figures, no “excellent / average / slow” bands and no percentile tables, because it has no real aggregated data to compute them from and will not invent them. Personal bests are kept locally in your browser for the same reason — the meaningful comparison is your run against your earlier runs.
For that comparison to say anything, hold the conditions still: the same keyboard, the same device, the same browser, and a similar state of rest and concentration. Then run the Typing Speed Test more than once. The platform-wide measurement rules behind all of this are on the methodology page.