Four governments publish guidance on the order in which Chinese characters are written. They do not merely disagree about characters. They disagree about what kind of thing the guidance is.
One publishes rules. One publishes answers and no rules. One publishes pictures and disclaims them. One publishes nothing at all and says so in the document where you would look.
Taiwan publishes the rules
Taiwan's education ministry issued a stroke-order handbook in 1996 covering 4,808 characters, and its second section is a list of seventeen numbered principles. They are specific enough to argue with. Rule 8 says a horizontal stroke that sits in the middle and projects on both sides is written last, and names 母 as an example. Rule 11 says that in characters containing 戈, the horizontal comes first and the dot and the downward-left stroke come last. Rule 16 covers shapes that cradle from below, and names 出.
Alongside the handbook there is an official animation site covering the same 4,808 characters, so every rule can be checked against a moving example.
The mainland publishes the answers
The current mainland standard is GF 0023 — 2020, issued in November 2020 and effective from 1 March 2021. It covers 8,105 characters — the whole of the general-use list — and supersedes the 1997 standard that most writing on this subject still cites.
The standard contains a foreword, scope, references, definitions, and drafting principles, but no section of general rules. The standard specifies the order for each of the 8,105 characters individually and does not state a single principle from which those orders follow.
Almost every account of Chinese stroke order gets this backwards. The familiar list of principles — horizontals before verticals, top to bottom, outside before inside — is taught universally and appears in every textbook. But that list is not the text of the mainland standard. It's pedagogy that summarizes the standard's output.
The standard is also explicit about who it is for. Its scope clause says it applies primarily to information processing, publishing and printing, and dictionary compilation, and may also be used in teaching and research. It does not present itself as binding on handwriting instruction.
Hong Kong publishes pictures, and tells you not to lean on them
Hong Kong's Education Bureau publishes a primary-school character and vocabulary list with stroke-order animations for 3,171 characters. Its own notes on use say the material — glyph shapes, stroke order and the rest — is for reference only and is not a sole or rigid standard, and the foreword adds that it must not be treated as a measure for tests or examinations.
Which standard the animations follow is not stated anywhere we could find. There is one Hong Kong document that did once prescribe stroke order, a 1986 character-form list produced in handwritten form with per-character orders — and when it was reset in 2007 the stroke-order content did not survive. As linguists at a Hong Kong university state flatly, Hong Kong and Macau have no officially promulgated stroke-order standard.
Japan publishes nothing, in writing
The obvious place to look for a Japanese standard is the 2010 general-use kanji list, a cabinet notification of 2,136 characters. A search of its official text for "stroke order" or "stroke count" finds zero occurrences of either.
That is not an oversight. A 2016 report from the national language subcommittee states explicitly that the general-use kanji list does not fix stroke counts or stroke order.
A 1958 handbook for teachers did cover the 881 characters then taught in primary school. But it was commercially published and disclaimed itself, stating that unlisted stroke orders were not necessarily wrong. It was published by a commercial house rather than the government printer.
What is binding today is the curriculum, which requires children to write "following stroke order" while naming no table to follow. Textbook screening criteria ask only that orders be the commonly accepted ones. The education minister told the Diet in 2016 that teachers had been urged toward flexible assessment.
One institution does mark it: the kanji aptitude test examines stroke order as a question type and names its own handbook as the marking authority. So the one place in Japan where stroke order is scored against a fixed list is a private examination.
How far apart the two standards actually are
Because Taiwan and the mainland both publish per-character orders, the two can be compared exactly. Researchers at a Hong Kong university did so across all 4,808 characters in Taiwan's standard.
The result: 1,692 characters — 35.19% — carry different stroke orders between the two standards. Their own summary is that on average fewer than three characters pass before the two disagree. Of those, 383 are cases where the printed shape is identical and only the order differs, and those 383 fall into just 32 groups — the first sixteen of which account for more than nine in ten.
The groups matter more than the count, because they are not scattered. They trace directly back to four of Taiwan's numbered rules. This is a disagreement between two published rule sets, not a collection of arbitrary choices.
Reading the orders as digits — 1 for a horizontal, 2 vertical, 3 left-falling, 4 dot, 5 turning — the disagreements look like this.
| Character | Mainland | Taiwan | What moves |
|---|---|---|---|
| 忄 | 442 | 424 | Mainland writes both dots then the vertical; Taiwan puts the vertical between them |
| 母 | 55414 | 55441 | The long middle horizontal: fourth, or last |
| 必 | 45434 | 45443 | The left-falling stroke: fourth, or last |
| 出 | 52252 | 25252 | Taiwan writes the long central vertical first |
| 皮 | 53254 | 35254 | Taiwan starts with the left-falling stroke |
| 戈 | 1534 | 1543 | Dot and left-falling stroke swap at the end |
| 里 | 2511211 | 2511121 | Inside the lower 土 |
One of those deserves a flag. The widely-read encyclopedia entry on this subject states that Taiwan writes 皮 with the turning stroke before the left-falling one, as the mainland does. Taiwan's own ministry animation begins with the left-falling stroke. The encyclopedia is wrong on that character.
The characters that do not differ
The claim that the standards "sometimes disagree" is vague. It's more useful to name characters where they are identical. All of these code the same way on both sides:
王 玉 主 青 田 由 右 左 有 布 方 火 心 女 州 臣.
Two of those are frequently offered as examples of divergence and are not. 方 is written identically in both. So are 右 and 左 — both jurisdictions write both of them horizontal-first. Those are differences between Chinese practice and Japanese practice, which is a different comparison and, as established above, a comparison against a convention rather than against a standard.
Stroke count is a different question
The standards treat order and count as separate terms with separate definitions, and they can disagree about either independently.
| Character | Mainland count | Taiwan count |
|---|---|---|
| 及 | 3 | 4 |
| 巨 | 4 | 5 |
| 瓦 | 4 | 5 |
These three are the cleanest examples because the printed shapes are effectively the same — the mainland merges strokes that Taiwan counts separately. A reader counting 及 and getting four is not making a mistake; they are using the other standard.
If you go looking for this in Unicode's character database, do not use its total-stroke field. It once carried two values for exactly this purpose and effectively no longer does, and its single value follows the traditional count. Compare that against a Taiwanese source and it will report agreement on precisely the characters that disagree.
Where Stroke Order Matters
Dictionary ordering is one example. A separate mainland standard defines a five-level cascade for sorting characters. The sequence begins with stroke count. For characters with equal counts, it proceeds to a stroke-by-stroke comparison (horizontal, vertical, left-falling, dot, turning), then to sub-forms of those strokes, then to how strokes connect — the level that finally separates 王 from 玉 — and last, to overall structure.
A change in a character's stroke order has a direct consequence: it moves the character's position in any dictionary or index built on that sorting rule. The 1997 mainland standard adjusted the orders of two characters and noted that their positions in the sort order moved with them.
The other place order is load-bearing is stroke-based input, where the five stroke types are mapped onto the number keys and a character is typed by entering its strokes in standard order. Get the order wrong there and you get the wrong character, with nothing to fall back on.
Where it does not decide anything, despite what you have read
The usual justification for learning stroke order is that handwriting input depends on it. For current systems, this turns out not to be true.
Image-based recognition cannot use stroke order at all — the order is not in the input. Apple's published account of its Chinese handwriting recogniser says it chose an image-based approach specifically for robustness toward non-standard stroke order, and its support documentation tells users plainly that stroke order does not matter. Google's paper on its production handwriting system describes an earlier generation that deliberately normalised writing order away.
Order does carry signal when a system is built to use it: one study found the same network reading stroke sequence scored 88.01% against 79.36% reading the image alone. But that measures what order adds when writers use the standard one. It does not measure what happens when they do not.
We could find no published research on that question. The one published benchmark comparing six commercial Chinese handwriting input methods varies the character set and does not vary stroke order at all — the phrase does not occur in it. The most candid vendor statement we found belongs to a dictionary app, which says its recogniser is fairly tolerant of stroke-order mistakes though more accurate with the correct order. That is an honest description of a soft dependency, and it is a long way from requiring one.
Which of the four applies to you
If you are teaching or learning, use the standard of the place whose examinations you will sit, and expect roughly one character in three to differ from the other side's.
If you are building a dictionary, an index or a stroke-input method, the order is load-bearing and you need to name which standard you have implemented.
If you are worried that your phone will not read your handwriting, that is the one thing on this list you can stop worrying about.
The documents behind this
The mainland standard is GF 0023 — 2020, read from the education ministry's own published PDF, along with its 1997 predecessor and the two 1999 standards for the larger character set and for sort order. Taiwan's rules and per-character orders are from the education ministry's 1996 handbook and its official animation site. Hong Kong's position is from the Education Bureau's own notes on use and foreword. Japan's is from the 2010 cabinet notification, searched directly, and the 2016 subcommittee report. The 35.19% figure and the group analysis are from a Hong Kong university comparison of all 4,808 characters. Individual stroke orders quoted here were read from the two governments' own published data rather than from secondary summaries. All were read on 20 September 2026.
Two limits. The Japanese character-by-character differences that circulate widely were not verified here against a Japanese primary source, and given that Japan publishes no normative table there may be nothing to verify them against — they are statements about convention. And 卍 appears in neither standard: it is not among the mainland's 8,105 nor Taiwan's 4,808, so any stroke order given for it is a convention with no standard behind it.
