To be conversationally fluent in French you need roughly 3,000 high-frequency word families, which covers about 95% of the words in everyday speech and writing. To read novels and newspapers without reaching for a dictionary, the research points to around 8,000-9,000 families (Nation, 2006). "Fluent" is not one number. It is a coverage target attached to a specific goal, and the gap between chatting comfortably and reading Camus unaided is several thousand words wide.
What counts as a word in French#
Before any number means anything, you have to know what is being counted. The same page of French yields four very different totals.
| Unit | Definition | French example |
|---|---|---|
| Token | Every running word, counted every time it appears | parle, parle, parlons = 3 tokens |
| Type | Each distinct written form, counted once | parle, parlons = 2 types |
| Lemma | A headword plus its inflected forms | parler covers parle, parles, parlons, parlait, parlerai |
| Word family | A lemma plus its transparent derivations | parler plus parole, parleur, reparler |
French inflection makes this gap large. A single regular verb has dozens of written forms across tense, person, and mood, and adjectives and nouns carry gender and number. One useful quirk works in your favor: many of those written forms are homophones. Parle, parles, and parlent are spelled differently but pronounced identically, so the number of forms you must distinguish by ear is much smaller than the number you must distinguish on the page. When a course promises "3,000 French words," it almost always means 3,000 lemmas or families. Lonsdale and Le Bras's A Frequency Dictionary of French (2009), a standard reference, is organized around the top 5,000 lemmas for exactly this reason.
The coverage thresholds behind "fluent"#
Vocabulary research does not measure fluency directly. It measures coverage, the percentage of running words in a text that a reader already knows. Two thresholds recur across the literature.
95% coverage is often described as the floor for adequate comprehension when a reader can guess the rest from context (Laufer, 1989). That is about one unknown word in twenty, or roughly one per line of a paperback. 98% coverage is the level Hu and Nation (2000) associated with independent reading, where the text carries itself and a dictionary becomes optional rather than necessary. That is about one unknown word in fifty.
Nation's 2006 analysis put the vocabulary needed for that 98% figure at roughly 8,000-9,000 word families for written text and somewhat less, around 6,000-7,000, for spoken language. Speech reuses a smaller core, which is why learners can hold a conversation long before they can read a novel comfortably.
How much text the top 1,000, 2,000, and 5,000 words cover#
Word frequency is extremely lopsided. A handful of items appear in nearly every sentence, and a long tail appears almost never. Zipf (1949) formalized the shape, and it is why your first thousand words buy so much and your eight-thousandth buys so little.
| Most frequent word families | Approx. text coverage | What it typically unlocks |
|---|---|---|
| 1,000 | ~80-85% | Survival French, the gist of simple conversation |
| 2,000 | ~90% | Everyday talk with gaps; graded readers |
| 3,000 | ~95% | Comfortable spoken fluency; most daily topics |
| 5,000 | ~96-97% | Simplified news, subtitled film with context |
| 8,000-9,000 | ~98% | Unedited novels and newspapers, read for pleasure |
The high-frequency band in French is dominated by function words and workhorse verbs: de, le, et, être, avoir, faire, aller, pouvoir, dire, vouloir. These few hundred items are unavoidable, which is exactly why learning them pays so well. The mirror image is the discouraging part. Climbing from 95% to 98% means acquiring several thousand progressively rarer words, each of which you will meet only occasionally. Diminishing returns are built into the distribution, not into your study method.
English speakers get an unusual head start here. Finkenstaedt and Wolff's 1973 survey of English etymology found that a very large share of the English lexicon came from French or Latin, which means thousands of French words are recognizable on sight. That advantage is real for reading and nearly worthless for listening, and it comes bundled with faux amis such as actuellement (currently, not actually) and librairie (bookshop, not library).
French vocabulary by CEFR level#
The Common European Framework describes what learners can do, not how many words they know. It specifies no vocabulary sizes at all. The mapping below combines Milton's vocabulary-size research (Milton, 2009; Milton and Alexiou, 2009) with published exam wordlists and should be read as an orientation, not a specification.
| CEFR level | Approx. word families | What it typically supports |
|---|---|---|
| A1 | 500-1,000 | Greetings, numbers, ordering food, simple personal facts |
| A2 | 1,000-2,000 | Routine exchanges, past and future tense, short simple texts |
| B1 | 2,000-3,000 | Handling travel and work situations, following clear speech, graded readers |
| B2 | 4,000-5,000 | Discussing abstract topics, reading the press with occasional lookups |
| C1 | 6,000-8,000 | Fluent professional and academic use, most literature |
| C2 | 9,000-10,000+ | Near-native range, including idiom, register, and specialist prose |
Two learners at the same level can differ by thousands of words depending on what they read, so treat these bands as overlapping clouds rather than lines. The same shape holds across related languages, which is why the equivalent Spanish vocabulary targets land in almost the same place.
Why every number here is an estimate#
Be skeptical of anyone who quotes these figures to three significant digits. The variation comes from four places. The unit matters most: counting families rather than lemmas can cut a total by a third or more. The corpus matters next, since a frequency list built from subtitles, such as those derived from the OpenSubtitles collection, weights conversational vocabulary very differently from one built on newspapers or literary prose. The definition of "knowing" a word matters too, because recognizing chuchoter in a sentence is far easier than producing it on demand, and every learner's passive vocabulary outruns the active one. Finally, coverage is not comprehension. You can know 98% of the words in a French legal contract and still miss the point, because specialist domains carry their own vocabulary that frequency lists never reach.
How to actually build a French vocabulary that size#
Word lists alone are slow, and they decay. Two mechanisms with real evidence behind them work far better in combination.
Volume of comprehensible reading. Krashen's Input Hypothesis (1985) holds that we acquire language by understanding messages slightly above our current level. Reading French you can mostly follow exposes you to high-frequency words dozens or hundreds of times in natural context, which is what turns a recognized form into a known family. If you want the mechanics, comprehensible input explained covers the argument and its limits.
Scheduled review. Reviewing a word at expanding intervals fights the forgetting curve Ebbinghaus documented in 1885 far more efficiently than rereading. The research behind spaced repetition is one of the most replicated findings in learning science, confirmed across dozens of studies by Cepeda and colleagues in 2006.
LingoBlend is built on that combination. Paste a French article, a chapter, or a recipe, set the blend slider, and the app weaves target-language words into text you already understand, which is the diglot weave technique from Burling (1968) with a modern engine behind it. Tapping a blended word shows its meaning, its grammar, and its base form, so voudrais maps back to vouloir and you build families instead of collecting orphan forms. Saved words then enter an SM-2 spaced-repetition queue and feed five review games. The French learning guide is a reasonable place to start planning.