Measure the four skills separately and expect the answer to be lopsided. Reading and writing run at your own pace, so you can test them alone and honestly. Listening and speaking run at somebody else's pace, so testing them requires either a recording or another person, which is why almost nobody does it and why almost every self-estimate is too high.
Which skills you over-rate is not random, and it has nothing to do with modesty or bravado. It follows directly from which skills you can observe yourself performing, and it points the same way for nearly everyone.
Eight ways to find out where you actually are
- recognition overstates your level, often by a whole band, because multiple choice tests what you can pick rather than what you can produce.
- a free placement test gives a range, which is enough to choose material and not enough to put on an application.
- can do statements are the real check. Read the list for a level and ask honestly whether you can do each one today.
- test production separately. Your speaking level and your reading level can differ by two bands and usually do.
- record yourself for two minutes on an unprepared topic and listen back, which is uncomfortable and accurate.
- read something unadapted, a newspaper column, and count the words you had to look up per hundred.
- a mock exam is the honest measure, timed and complete, if a score is what you need.
- test again in three months, the same way, because a level is only meaningful as a direction.
Four tests, each taking about twenty minutes
For reading, take a page of unsimplified text of the kind you actually want to read and count the words you have to stop for. That number, per page, is the measurement, and it is more useful than any band because it tells you directly whether the material is available to you today.
For listening, find something unscripted, with an accent you have not been trained on and no subtitles, play three minutes once, and write down what happened. Once, at full speed, with no replay, because a replay measures something you will never get.
For writing, produce four hundred words in forty minutes on something you have an opinion about, then have somebody else find the errors. Counting your own is not a measurement: the errors you can find are by definition the ones you already know about.
For speaking, record two minutes about your own week and listen to the whole recording. That last instruction is the one people skip and it is the entire test.
Where it stops: one test, called a level
The stall is taking a single measurement and treating it as the answer. An online test is sat, a number comes back, and it becomes the level, which is then repeated for years in conversations and on forms.
Almost every free test measures reading and listening, because those are the two that can be scored automatically, and reading is usually the strongest of the four. So the single number is close to the top of your range rather than in the middle of it, and the skill somebody will actually experience when they meet you is one or two bands below it.
Nothing corrects it, because nobody ever tells you what level you sounded like. The number stands for years, and the mismatch only surfaces at an interview or in a meeting where the gap between the claim and the performance is visible to everyone except the person who made the claim.
A free estimate, and what it is estimating
Test Your Vocab is free, needs no account and takes a few minutes: you tick the words you know from a spread of frequencies and it returns an estimated vocabulary size. As a number it is genuinely informative about where you sit on the supply side.
It replaced guessing at your own vocabulary size, which people do badly in both directions and with great confidence. Free.
Its limitation is the whole argument of this page in one instrument. It measures recognition of written words, which is the single skill that runs furthest ahead of the others, so a large result confirms the half of you that was already strongest. Use it as one of the four numbers rather than as the answer, and pair it with the recording, which measures the half nobody wants to look at.
What it looks like when it goes wrong
The recognisable failure is a candidate who tests at one band and performs two lower in the room. Nothing was faked. The test measured the skills that wait for you and the room did not wait.
The second failure is refusing to listen to the recording. People record themselves willingly and then do not play it back, which converts the only cheap measurement of the weakest skill into a file nobody opens.
What the bands the results get reported in actually describe, and why none of them names a structure, is what A1 through C2 mean in things you can do, and the reason a self-estimate can sit still for three years without anybody querying it is why progress stalls at B1, and what moves it.
The question this leaves is whether you need a measurement at all or a number somebody else will accept, because those are different products with different prices, and only one of them can be got for nothing tonight: what each test scores, and how to prepare for yours.