HTML Entities Reference

How to use it

1. Paste the character into the lookup box at the top. A pasted symbol, a code point written U+2014, 0x2014 or 8212, or a reference such as — all resolve to the same answer, and each spelling has its own copy button.
2. Filter the table when hunting rather than checking: type part of a Unicode name (dagger), part of an entity name (rarr), or the character. The eleven group chips — reserved, spaces, typography, currency, math, arrows, greek, cyrillic, latin, shapes, symbols — narrow it further; press / to jump to the box.
3. Read the row, not just the glyph: it prints the Unicode name, the decimal and hex spellings, every other name for that character, and a badge when the name is one of the 106 that decode without a semicolon.
4. Link straight to a row: every row is anchored by its entity name, so /html-entities#mdash opens on that entry.

You have a character and you need its HTML code. Paste the symbol itself — or a code point like U+2014 — into the lookup and get its named, decimal and hex spellings back. The lookup knows all 2,125 named references; the table below prints the 252 HTML 4.01 defines, server-rendered so find-in-page works before any script runs. Nothing you type is sent anywhere.

How it works

The names are older than the web. SGML shipped standard public entity sets — ISOlat1 for accented letters, ISOnum for signs, ISOpub for publishing marks — and early HTML adopted them, which is why lt, amp and eacute read like typesetting abbreviations. HTML 4.01 formalised three of those sets and stopped at 252 names — the set the table below prints. The WHATWG list took over, absorbing MathML's mathematical vocabulary: of its 2,125 names, 1,033 are math and 308 arrows.

There are more names than there are characters. The 2,125 references cover 1,511 distinct characters, so 1,051 of them carry at least one alias — a casing pair kept for old documents (AMP beside amp), or one symbol arriving from two vocabularies, as when the approximation sign answers to six names, TildeTilde and thickapprox among them. A printed row lists the rest under Also spelled, and filtering finds them.

One hundred and six of the entries are decoded even when the semicolon is missing, and the table badges each. They are a frozen compatibility inheritance: nothing new may join, which is why apos — a latecomer from XML — needs its semicolon while quot does not. Write the semicolon regardless. Where a document is already missing one, the browser parsing it is the final authority.

The table is generated, not curated. A committed script reads the WHATWG entities.json snapshot dated 2026-07-29 and asks Python's unicodedata module for each character's official Unicode name, so the label beside a glyph is the character database's own wording; Tab and NewLine take theirs from the Unicode NameAliases file, control characters having no Name property. To convert a document instead, use /html-decode.

Examples

—

— · — · — — EM DASH

U+00A0

    (legacy) ·   ·   — NO-BREAK SPACE
2 names spell this character; they are equivalent.
HTML also decodes this name without the semicolon, for historical reasons. Write the semicolon.

≂̸

≂̸ ≂̸ · ≂̸ · ≂̸ — MINUS TILDE + COMBINING LONG SOLIDUS OVERLAY
This reference stands for 2 code points, not one character.
2 names spell this character; they are equivalent.

😀

😀 · 😀
This character has no named reference in HTML — use the decimal or hex spelling.

rightwards arrow

105 rows match — first: → · → · → — RIGHTWARDS ARROW

Every example on this page is executed against the same engine that powers the tool, on every build.

Common errors

No reference matches that filter

The filter state where a query matches none of the printed rows — usually a description the character database does not use, since a row's searchable text is its entity names, its Unicode name, its group, its code point and the character itself. The lookup above always answers.

Look up 64 characters at a time — this page finds codes for characters, it does not convert documents.

A typed refusal past 64 code points: a lookup takes one character, a filter a few words. Pasting a paragraph means you want /html-decode instead. Nothing is truncated here: a partial answer about a partial paste is worse than a refusal.

That is not a Unicode code point — try a character, a name like mdash, or U+2014.

What a number Unicode cannot hold returns: anything above U+10FFFF, or a surrogate value in D800–DFFF. Real characters are never refused — an emoji of several code points is reported piece by piece.

Keyboard shortcuts

/

Focus the filter box