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
Self-assessment audit
Eight principles for establishing an honest language baseline
Free online tests overstate your level by measuring recognition rather than production. A genuine assessment isolates what you can produce under pressure.
Diagnostic evaluation
- YesSeparate productive skills from receptive recognitionSpeaking and reading typically differ by two full CEFR bandsYes
- YesAudit functional can-do statements directlyDetermine whether you can perform specific real-world tasks todayYes
- YesRecord and listen to unprepared spoken speechTwo minutes on an unscripted topic reveals your actual production baselineYes
- YesCount lookups in unadapted textsTracking unknown words per page in a newspaper gives a concrete reading metricYes
- NoRely on multiple-choice placement quizzes as an absolute levelMultiple choice tests what you can recognize, which inflates your true bandNo
A placement test gives a general range for choosing material, but only live production testing reveals your true operational band.
- 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.
Why the estimate is high
Two axes, and the two skills you can test alone are the two that ran ahead
The bias in every self-assessment falls out of this grid rather than out of anybody's temperament.
you can score it alone
- the easiest to measure
- and usually the highest of the four
- three minutes, unscripted, no replay
you need another person
- four hundred words, forty minutes
- counting your own errors is not a measurement
- two minutes recorded
- and then actually played back
The two cells on the left are the ones every free online test scores, and they are the two that were already ahead. What the grid does not say is how far apart the four usually are, which is more than most learners expect.
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.
The single-score trap
What automated tests measure compared to what live encounters demand
Taking a single online test and treating it as your definitive level creates a false sense of security that collapses in real meetings.
- What automated online tests measureEasy to score, self-paced, and heavily weighted toward your strongest skill
- Reading comprehension with unlimited time to reread
- Selecting from multiple-choice answer options
- Clean audio recorded by professional voice actors
- Instant automated score matching your peak receptive capability
- What live workplace communication demandsUnscripted, rapid, and unforgiving of production gaps
- Formulating spontaneous sentences with zero delay
- Decoding unfamiliar regional accents in noisy rooms
- Repairing conversational breakdowns on the fly
- Exposing production skills that lag two bands behind reading
Assessment distortion
Online tests score the skills that wait for you. The room you walk into does not wait, which is why candidates test at one band and perform two lower.
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.
Supply-side metric
Anatomy of the Test Your Vocab statistical estimator
Test Your Vocab offers a fast, free estimate of passive vocabulary size, but measuring what you recognize confirms the half of you that was already strongest.
- 01Test Your Vocab
- 03Estimated total vocabulary count
- 04Replaces ungrounded guesswork
- 05Recognition runs ahead of active speech
- 01Assessment ToolStatistical vocabulary size estimator
- 02MethodologyTick words you recognize across progressively rarer frequency tiers
- 03Output ResultA helpful quantitative measure of passive lexical inventory
- 04Core ValueGives an objective picture of supply-side word recognition
- 05Key LimitationConfirms the strongest receptive skill while telling nothing about spoken fluency
- 06Necessary CounterpartMust be paired with audio playback to measure active production
Use vocabulary estimation as one metric among four. To know what you can actually produce, record your voice and listen back.
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 test gap
The test environment compared to the live conversational room
When candidates test at one band and perform two bands lower in real life, nothing was faked: the test measured skills that wait, and the room did not wait.
The mismatch between test scores and live performance stems from relying on passive measurements. The room tests active retrieval under pressure.
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.