Where text breaks

49 classes · 217,414 characters

The Unicode line-breaking classes, counted by character.

The finding. Every design system assumes a word is a run of letters between spaces. That describes one character in eight. Ideographic characters — which break almost anywhere — outnumber alphabetic ones 6.4 to 1, and an entire class cannot be broken at all without a dictionary.

breaks anywhere 84.7%

breaks spaces 12.9%

breaks attaches 1.2%

breaks other 0.6%

breaks locale 0.4%

breaks dictionary 0.3%

Filter by class or name

What an engine must know

The two-letter codes are the specification's, and they are not guessable. ID reads as identifier and means ideographic; NS is nonstarter, a character that may not begin a line; PR and PO are the symbols that sit before and after a number and must never be separated from it. Every class carries its name, an example you can look at, and what it forces a layout engine to know.

Class	Name	Looks like	Characters	What it means
ID	Ideographic	漢字仮	172,561	Breaks between almost any two characters. No spaces required, and none used.
XX	Unknown	—	137,468	Unassigned codepoints, taking the default behaviour. The largest class by count, and it is empty of actual characters.
AL	Alphabetic	A b C	26,954	Breaks at spaces — the Latin model, and the one every design system assumes.
H3	Hangul LVT Syllable	각 난	10,773	A Korean syllable block with a trailing consonant.
CM	Combining Mark	◌́ ◌̈	2,512	Takes the class of the character it attaches to. An accent has no line-breaking behaviour of its own.
SG	Surrogate	—	2,048	Half of a code unit pair. Not a character on its own, and never appears in well-formed text.
SA	Complex Context Dependent	ก ข ค	757	Cannot be broken without a dictionary or a trained model — Thai, Lao, Khmer, Myanmar. There is no space and no character that marks the word edge.
AI	Ambiguous	± × ÷	718	Resolved by locale, not by the character. The same codepoint breaks differently in two languages.
NU	Numeric	0 1 2	705	Digits, which must not be split from their separators.
H2	Hangul LV Syllable	가 나	399	A Korean syllable block of lead consonant plus vowel.
AK	Aksara	क ख	329	A Brahmic orthographic syllable. Breaking inside one is wrong, and finding its edge takes cluster analysis rather than a lookup.
BA	Break After	– ⁄	263	A break opportunity follows this character.
AS	Aksara Start	ক খ	214	The consonant that begins a Brahmic orthographic syllable.
JT	Hangul T Jamo	ᆨ ᆩ	137	A Korean trailing consonant, as a separate codepoint.
EB	Emoji Base	👍 🙌	134	An emoji that a skin-tone modifier can attach to.
JL	Hangul L Jamo	ᄀ ᄁ	125	A Korean leading consonant, as a separate codepoint.
OP	Open Punctuation	( [ 「	95	An opening bracket or quote. Nothing may break after it — it must stay with what it opens.
JV	Hangul V Jamo	ᅡ ᅢ	95	A Korean vowel, as a separate codepoint.
CL	Close Punctuation	) ] 。、	94	A closing bracket, or the CJK full stop and comma. Nothing may break before it.
HL	Hebrew Letter	א ב ג	75	Hebrew letters. Split from Alphabetic because a hyphen behaves differently after them — Hebrew does not break at a maqaf the way Latin breaks at a hyphen.
PR	Prefix Numeric	£ $ € +	67	A symbol that comes before a number and must not be separated from it — a currency sign, a plus, a minus.
CJ	Conditional Japanese Starter	ぁァー	60	May or may not start a line depending on Japanese typographic preference — strict or normal.
BB	Break Before	´ ˈ	55	A break opportunity precedes this character.
GL	Non-breaking Glue		41	Forbids a break on both sides — the non-breaking space, and what holds “10 km” together.
EX	Exclamation / Interrogation	! ? ！	40	Nothing may break before it. The mark belongs to the sentence it ends.
QU	Quotation	" ' «	39	A quote mark that may open or close, so the engine cannot tell from the character which side needs protecting.
PO	Postfix Numeric	% ° ℃	38	A symbol that comes after a number and must not be separated from it — a percent sign, a degree, a unit.
NS	Nonstarter	々ゝヽ	37	Cannot begin a line. Japanese small kana and iteration marks belong to the character before them, so a break in front of one is a typographic error.
RI	Regional Indicator	🇬🇧	26	Letters that only mean anything in pairs — two of them make a flag, and a break between them makes two letters.
HH	Unambiguous Hyphen	‐	11	The real hyphen character, added because the ASCII one could not be read reliably.
IS	Infix Numeric Separator	. , : ;	10	Sits inside a number and must not be split from it — the decimal point in 3.14, the comma in 1,000.
VI	Virama	◌् ◌্	7	The mark that removes a Brahmic consonant's inherent vowel and joins it to the next.
CP	Close Parenthesis	) ]	6	Split from Close Punctuation so that a letter may follow it — the case of (s) in “form(s)”.
IN	Inseparable	‥ …	6	A run that must not be broken internally — the dots of an ellipsis or a table-of-contents leader.
AP	Aksara Pre-Base	◌ெ ◌െ	6	A Brahmic vowel sign written before the consonant it is pronounced after.
EM	Emoji Modifier	✋🏻 ✋🏿	5	A skin-tone modifier, which never separates from the emoji it follows.
BK	Mandatory Break	U+000C U+2028	4	Always breaks. The line ends here whatever else is true.
B2	Break Before and After	—	3	A break may fall on either side, but not between two of them.
VF	Virama Final	◌്	2	A virama at the end of a cluster, which does not join anything to it.
WJ	Word Joiner	U+2060	2	Invisible, and forbids a break. The inverse of the zero width space.
LF	Line Feed	U+000A	1	Always breaks.
CR	Carriage Return	U+000D	1	Always breaks, and pairs with a line feed as one break rather than two.
SP	Space	U+0020	1	The break opportunity itself, and the only one most software knows about.
HY	Hyphen	-	1	The ASCII hyphen-minus, which is a hyphen, a minus sign and a dash depending on context — so the engine has to guess.
SY	Symbols Allowing Break After	/	1	The solidus. A break may follow it, which is why a long URL can wrap after a slash.
NL	Next Line	U+0085	1	Always breaks. An EBCDIC inheritance that survives in Unicode.
ZW	Zero Width Space	U+200B	1	An invisible break opportunity. Nothing is drawn, and the line may end there.
ZWJ	Zero Width Joiner	U+200D	1	Invisible. Binds what is either side of it into one unit — how multi-person and profession emoji are built.
CB	Contingent Break	U+FFFC	1	The character stands in for an embedded object, and the object decides. An inline image, in other words.

Forty-nine classes, and one of them is the one every design system implements. “Breaks at spaces” covers 12.9% of assigned characters. The rest of the table is what a text layout engine has to do after the easy case runs out — and what a component library that measures a string by counting spaces silently gets wrong.

Source. Unicode Character Database, LineBreak.txt (UAX #14) · Unicode Licence · retrieved 2026-07-31. This collection is reproduced.

The documents. UAX #14 Line Breaking · Unicode Character Database

The building

The desk

Where text breaks