About a thousand Chinese characters make up 90% of written Chinese. We counted three large text collections, and the 90% mark came at 1,057 characters in a collection of books and articles, 1,069 in Chinese Wikipedia and just 717 in film subtitles, where everyday speech leans on fewer characters. What you read decides the answer more than any rule of thumb.
This guide sets the three counts side by side, explains why a character count is not the same as a reading level, and shows where the familiar figure of 3,500 characters comes from. Our character frequency tool lists the 200 most frequent characters in Chinese Wikipedia with their cumulative coverage.
Three collections of text, counted
Each collection is a corpus, a large body of text in which every character has been counted.
- Chinese Wikipedia, counted by us: every article in the dump of 1 September 2026, about 776 million characters, with templates and reference lists removed and traditional characters converted to simplified before counting.
- Jun Da's Modern Chinese list, compiled at Middle Tennessee State University from about 194 million characters of fiction and non-fiction.
- SUBTLEX-CH, a count of about 47 million characters of film subtitles, published by Qing Cai and Marc Brysbaert in 2010.
| Share of text covered | Wikipedia | Books and articles (Jun Da) | Subtitles (SUBTLEX-CH) |
|---|---|---|---|
| 50% | 191 characters | 152 | 62 |
| 80% | 652 | 611 | 356 |
| 90% | 1,069 | 1,057 | 717 |
| 95% | 1,525 | 1,566 | 1,111 |
| 99% | 2,795 | 2,838 | 2,076 |
The two written collections agree closely from about the thousandth character onwards. Subtitles are the outlier all the way down: sixty-two characters cover half of what is said on screen.
Why the subtitles need so few
Spoken language repeats itself. In the subtitles, the ten most frequent characters make up 23.1% of all text; in the books and articles, 13.9%; in Wikipedia, 10.4%.
The clearest single example is 我, "I". It is the most frequent character in the subtitles, ninth in the books and articles, and 465th in Wikipedia. An encyclopedia is written in the third person and is full of dates and places, so its second most frequent character is 年, "year", and 国 (country), 区 (district) and 市 (city) rank far higher there than in either of the others.
Where 3,500 comes from
The figure most learners meet is 3,500, and it comes from the List of Frequently Used Characters in Modern Chinese (现代汉语常用字表), published in 1988: 2,500 common characters and 1,000 less-common ones. Its compilers tested it on two million characters of 1987 text, mostly from the People's Daily, and reported that the 2,500 common characters covered 97.97% of it and the next 1,000 another 1.51%, a total of 99.48%.
Our counts agree with that test. In our analysis, the 3,500 characters from the 1988 list cover 99.2% of both the books-and-articles collection and the subtitles. Across all three of our collections, the 3,500 most frequent characters cover at least 99.5% of the text. The 2013 table that replaced the 1988 list, the Table of General Standard Chinese Characters, keeps 3,500 characters as its first level.
Covering 90% of characters is not understanding 90%
A coverage figure counts characters, but reading happens at the level of words. Knowing 90% of the characters in a text still leaves one in ten unknown, and a familiar character can appear in a word whose meaning is not obvious from its parts. Research on reading in a second language suggests that comfortable independent reading requires knowing about 98% of a text's words—a much higher bar than 90% of its characters.
The main use of these coverage figures is as a guide to learning order. The first few hundred characters clearly repay study far more than the next few thousand. In Wikipedia, the first 500 cover 73.6% of the text; the next 3,000 raise that to 99.5%.
What these counts do not cover
All three collections are counted in simplified characters. The books-and-articles and subtitle collections draw on mainland Chinese text; Chinese Wikipedia is written by contributors in many places, and about half of it is stored in traditional characters, which we converted before counting. A reader of traditional characters, or of Chinese written in Taiwan, Hong Kong, Singapore or Malaysia, should not assume the numbers carry over exactly. The shape of the curve — a few hundred characters doing most of the work — is very likely to hold; the particular ranks will not.
Our tool's figures changed on 21 September 2026. The table it showed before was credited to Jun Da's list, but most of its coverage figures did not match that list, and the list's terms require permission for commercial use. The tool now shows our own Wikipedia count.
How the counts were made
The Wikipedia figures are our own count, described above; the table in our tool is offered under the same CC BY-SA 4.0 licence as Wikipedia's text. The Jun Da and SUBTLEX-CH figures were computed from the character lists their authors publish, by adding up each character's share of the text in rank order; we cite the resulting figures rather than reproduce the lists. The 1988 figures are from the list's published explanatory notes. Coverage is cumulative. The figure for 90% coverage, for example, is the number of unique characters, starting from the most frequent, whose total occurrences add up to 90% of all characters in the collection.
View author profile → · Editorial policy
