Guide to using special characters in HTML


Entering characters

You can enter a character in an HTML document in the following ways:

In addition to these primary methods, a character might be entered via JavaScript code, where a string literal can contain a character using a “backslash escape” such as \xe9 or \u00e9. In CSS code, character data to be added to the document rendering might use a different “backslash escape,” such as \0000e9.

Use UTF-8 if possible

<!doctype html>
<meta charset=utf-8>
Voilà ! ☺☻

Usually the best option is to enter characters as such, using a Unicode-capable authoring tool and saving and serving the document as UTF-8 encoded. This makes the HTML source more readable and avoids steps that might introduce errors. Just make sure that UTF-8 is properly declared:

This page does not use UTF-8, because it would complicate maintenance. I edit the page via a Unix shell connection using the Emacs editor.

Dealing with font problems

The problems of representing characters in HTML have generally been solved thanks to better browsers. The problem of rendering characters remains.

By the specifications, browsers should display a character if there is any font in the system that contains it. If the fonts specified by the author (in CSS font-family settings or, rarely these days, using font markup in HTML) do not contain the character, browsers are sup­posed to use fallback fonts. The same applies if no fonts are specified by the author; brows­ers should use primarily their default fonts, using alternate fonts for any character not covered by the primary font.

In practice, things don’t always work that way. Especially IE is notorious for its failures in this respect. It often fails to display a character, even though it could do that if it used all the fonts in the system. If a browser cannot render a character, it may show a small rectangle, possibly containing a question mark, ?, or some similar indicator. Here’s a quick test (char­ac­ter U+0840, which is probably not supported by any font on your computer): ࡀ.

The fallback font principle has pitfalls. The problem is that there is no guarantee that characters from different fonts fit together. In the the following screenshot, some of the Chinese characters disturbingly look bold. On a closer look, they turn out to be from a sans-serif font that differs from the basic font of the text. In any case, the appearance of the text is disturbing, and the reader may suspect that it is erroneous and not just poorly dis­played. This may happen if you have Chinese text and you do not specify the font, or you specify fonts that do not contain all the characters needed.

(Chinese text where some characters stand out as if they were in bold face.)

For texts in Latin letters, such problems have become relatively rare. Most commonly used fonts cover the Latin letters rather well. But if you either need some rare letters or use a special font, problems may occur. Fonts designed for special use, like a company’s brand font or an artist’s personal font, often have a limited character repertoire. When a fallback font is then applied, an accented letter like ă might appear stylistically quite different from the corresponding unaccented letter (a).

To conclude, you should

If you need to resort to fallback fonts, try to make their use consistent. For example, the Arial Unicode MS font is a suitable fallback for Arial and maybe for some other fonts in the same general design. But for an essentially different basic font, try to find fallback fonts that resemble it better.

There is handy information about font support for characters in the FileFormat.Info site, section Characters. The information does not cover all fonts, but it lists all the fonts that you can normally expect your visitors to have, and a lot more. The following simple form is a quick way of accessing the information:

Case study

Suppose that we wish to use the HOURGLASS U+231B character. According to FileFormat.Info, there are eight fonts that contain it. But if you look at the character in the fonts (you can click on the “View All” link on the result page to see them), you will notice that some of them are unsuitable. Moreover, as you check from Alan Wood’s font information page, most of the fonts are rather special. More exactly, this is the situation with them:

Arial Unicode MS
This is the largest font that most people have in their computers. But it is not always available: it is a commercial Microsoft product, shipped with Office 2000 and newer.
A very large font, freeware. Needs to be downloaded and installed.
Everson Mono
A freely downloadable monospace font, with a small license fee.
Nominally, an oblique (slanted) version of EversonMono, but there is no difference for a character like this. Moreover, in CSS, it is not to be referred to by its own name but by using Everson Mono and font-style: oblique or font-style: italic.
A generic fallback font, not suitable for any normal use, but can be used in testing.
A relative new free font, not much known.
Another relative new free font, not much known.
Unicode BMP Fallback SIL
Another generic fallback font, showing just the code number in a box.

To summarize, if you use the HOURGLASS character, visitors will see it if and only if they have Office in their system or they have some of the special fonts that one needs to download and install separately. Others will see just a question mark in a box, the code number 213B in a box, or something similar. If you decide to take your chances, then you should use a style sheet that applies the following declaration to an element containing the character:
font-family: Quivira, "Arial Unicode MS", Symbola, "Everson Mono";
The order of the fonts in the list is up to you of course, putting first those fonts where the rendering looks best.

Simple tests:
⌛ (HOURGLASS character with no font settings)
(HOURGLASS character with the above-mentioned font settings).

A mathematical case

Font support for CONTOUR INTEGRAL (U+222E) is relatively widespread, though the fonts are special fonts rather than commonly used copy text fonts.

The combined availability of the fonts on computers is fairly good, so you might not even need to do any font settings. However, it may still be useful to do such settings, partly because the same settings can be used for many other mathematical characters. In addition to potentially helping some browsers to render the character at all, the font settings may help to create more pleasant and uniform typographic appearance.

The main problem is the ordering of the fonts in a preference order. Normally, mathematical texts work better when rendered in serif fonts, since the serifs help to distinguish math­e­mat­i­cal symbols from each other and from other characters. Many symbols do not contain serifs but they may still match serif font design better. On the other hand, mathematical symbols should match their traditional shapes in books, if possible. Such considerations, as well as elimination of some redundancies, could lead to the following list, based on FileFormat.Info data on font support for CONTOUR INTEGRAL and some additional information:
font-family: "Arial Unicode MS", "Lucida Sans Unicode", "Cambria Math", "Asana Math", "OpenSymbol", "Symbola", "Quivira", "STIX", "Code2000", "DejaVu Sans", "unifont";

There is more information about using mathematical characters in HTML on the page Math in HTML (and CSS). See also Mathematical symbols in ISO 80000-2 – a test page.

Favor precomposed characters

Letters with a diacritic mark, such as “é”, have two representations:

If you are using character references, you would use &#xe9; in the first case, e&#x301; in the latter. Here is how your browser renders them in this context: é versus é.

The visual rendering is generally expected to be the same, but in reality it may vary. Sometimes the font being used does not contain the precomposed character. More often, the font does not contain the combining mark. When the mark is taken from a different font, disaster may result.

In most cases, precomposed characters work better. It is better to use a glyph designed by a typographer than to let browsers construct a character from a base character and a combining diacritic mark. Especially older browsers do the construction work often poorly. For acceptable rendering, the browser needs to inspect the properties of the base character and position the diacritic mark accordingly. For example, the diacritic mark in É needs to be placed in a vertically higher position than in é. (On the other hand, an advanced browser may recognize that the base character and the combining mark are equivalent to a precomposed character and use a glyph for it.)

Browsers, search engines, and other software (e.g., editors, when working on text copied from a web page) generally work better with precomposed characters, due to their simplicity. It’s just yet another character, one Unicode code point. Moreover, precomposed characters are used much more often than combining marks, so there is strong motivation for software designers to handle them.

However, not all combinations of letters and combining marks exist as precomposed characters. There is a fixed number of precomposed characters in Unicode. So for some (relatively rare) characters used in different languages and notations, you may need to use combining marks.

Help on selecting fonts

Italics and bolding

Be extra careful if the special characters might appear in italics or in bold. Most fonts with very large or specialized character repertoire lack italic and bold versions. Web browsers will still apply italic and bolding to them, but by algorithmic methods that produce very poor results. The slanting tends to be excessive.

In the image on the right, the letters “ag” appears first in Cambria as normal and italic, then in Arial Unicode MS as normal and (fake) italic. The letters “ag” nor­mal­ly have rather different shapes in italic; that belongs to the idea of italic. When they are set in Arial Unicode MS, which lacks italic, you can still apply e.g. <i> to them, but browsers will mechanically slant normal letters.

When applying “fake italic” and “fake bolding”, browsers may italicize or bold symbols that get badly distorted in such processes—think about slanting the vertical bar “|”. Even if the character remains recognizable, it often gets a wrong shape. Italic is supposed to pick up a particular design for letters, not to slant any character. There are two ways to prevent wrong effects:

Character vs. style difference

In many cases, the difference between two symbols in text can be regarded either as a char­ac­ter difference or as a difference in rendering style for a single character. For example, there are many characters in Unicode that are italic or bold counterparts of normal characters.

Mostly, it is best to use styling to make the difference. For example, we show an italic x, x, using suitable HTML markup (e.g., the i element), instead of looking for italic x as a separate char­ac­ter.

For example, if you need Fraktur (Blackletter) letters e.g. in mathematical texts, it is probably best to use normal letters and style them. You would use markup like <span class=frak>z</span> and a suitable CSS rule like .frak { font-family: UnifrakturMaguntia } to produce a Fraktur z, z. (This example uses a Google font.)

A different approach (which might be regarded as theoretically better, but practically not so) is to characters like U+1D537 MATHEMATICAL FRAKTUR SMALL Z. Since this character is present in few fonts only, and none of them is probably installed in your visitors’ computers, you would need to use a downloadable font in this case, too, in practice. This would mean using a large font like Quivira or Symbola.

Line spacing problems

When characters from different fonts are mixed, line spacing may become uneven. A drastic example is that if some char­ac­ters are taken from the Cambria Math font, there will be a huge amount of space above and below any line that contains them.

For example, the diameter sign (⌀) and the inch sign (″) are needed relatively often. However, they are not present in most fonts. Authors may thus encounter the problem when they try to use these characters instead of the common but logically and typographically inferior replacements like letter o with stroke (ø) and Ascii quotation mark (").

The reason is that fonts have different default line heights. The line height is the distance between baselines of text. The actual height of a line is determined by the maximum line height of its characters.

For example, for text in Cambria font in 16px size, the default line height is typically 19px (depending on browser). For Arial Unicode MS in the same size, the default line height is 21px, which causes some disturbance. But with Cambria Math, the default line height is 89px, causing a gross effect.

A simple fix is to set the line-height property in CSS. Its value can be specified as a pure number, which is interpreted as relative to the font size of the element. Typical defaults are around 1.2, but you might wish to use a larger value like 1.3 especially if there are tall characters in the text. The important thing is to level out the differences within text. Sample style sheet:
* { line-height: 1.25; }

Web fonts

Web fonts, i.e. fonts that are automatically downloaded from a web server to a client, can solve problems with fonts. One of the problems is that to cover most browsers, you would need to make a font available in different formats.

You can use tools like Font Squirrel @font-face generator, which generates the font files you need and a CSS file for using them. It has “Expert” option in its user interface, for selecting e.g. a collection of Unicode numbers or ranges of numbers, to avoid embedding a large font in its entirety. However, the generator is slow and fails for some fonts (just does not produce anything).

Another service is CodeAndMore, which is fast and works even for some fonts that Font Squirrel cannot handle. But it has nothing corresponding to the “Expert” option. For large fonts, the resulting files can thus be quite large.

However, Google Web Fonts offer a simple solution. They are currently suitable mainly for normal texts in different languages rather than mathematical, technical, or other special symbols.

For example, the Khmer (Cambodian) writing system is poorly supported in fonts. The odds are that a user does not see Khmer letters unless he has installed either Code2000 or some special, Khmer-oriented font. You can greatly improve the situation by choosing one of the Google Web Fonts that support Khmer, say Suwannaphum, and using it on your page as follows:
<link rel=stylesheet href="">
You would then use the font normally in CSS, e.g.
.khmer { font-family: Suwannaphum, Code2000; }
The Code2000 font is included to deal with the possibility that the user’s browser has been configured not to use web fonts. (It is not probable that Code2000 is available, but you can always try.)

The following line is Khmer text that has been set to use the Google Web Font:
ខ្ញុំអាចញ៉ាំកញ្ចក់បាន ដោយគ្មានបញ្ហា

Google has an activity that they call Early Access fonts, supporting several languages for which it can be difficult to find suitable fonts otherwise, such as Bengali, Georgian, and Kannada.

Characters as images

In the early days of the web, it was common to present special characters as images. Nowadays such methods are rarely needed, and you get much better rendering using characters. But sometimes you need to use a character that has too limited font support. You can then consider using an image. This can be a more practical choice than using a downloadable font, if you need just one character, or a few characters.

For example, if you need to use the character U+1F300 CYCLONE in text, you should be worried about fonts: only a few fonts like Quivira, Symbola, and Unifont contain it. But you can use 🌀 in text as an image:
<img src=cyclone.gif alt="&#x1f00;" class=char title= "the character “cyclone”"
if you have it in suitable size in cyclone.gif.


A simple way to get an image representing a character is to use Unicode Character Search at Each character description page contains an image of the character, in a sans-serif font, as a 100 by 100 pixels PNG image.

Fonts with proprietary encodings

Tricks that try to extend character repertoire with fonts in proprietary 8-bit encodings are still in use to some extent. The idea in them is to some collection of characters positions from 0 to 255 in a font and to persuade programs into using that font.

For example, the widely available Symbol font is such a font, and so are the Wingdings fonts. If you write <font face=Symbol>X</font> in HTML, then the correct things for a browser would be to show the letter X in some other font, because Symbol does not have letter X. This is how e.g. Firefox behaves. Instead, many browsers (like IE) render the character that has been placed into the code position of letter X (58 in hexadecimal), namely the Greek capital letter xi (Ξ).

Such tricks with the Symbol font were rather common in the early years of the Web, but for a long time, much more reliable alternatives have existed for any character that appears in Symbol.

However, for some writing systems the approach is still used quite a lot. If the amount of characters is relatively small and the characters are not well supported by commonly available fonts, the trick looks like a simple solution. Nowadays, sites that use it (such as Eenadu, a Telugu-language newpaper that uses the Eenadu font) often use the font as a web font (downloadable font), so most users will see the text as intended. If they try to copy and paste the Telugu text, they will notice that it turns to gibberish like “ª½Â¹h¢.”.

The trick works in this case on Firefox, too, with caveats—it does not work if browser settings somehow disallow the use of the particular font. The reason is that the Eenadu font does not contain information about its coverage for characters.

There are two documents that are still worth reading if considering the use of this trick: <FONT FACE> considered harmful by Alis Technologies and Using FONT FACE to extend repertoire? by Alan J. Flavell.

Tooltips and other special texts

In addition to normal content in a document, browsers may render other texts in special ways. This includes

The font(s) used in tooltips depend(s) on the browser, which may or may not use settings made at the operating system level. Thus it may be controllable by the user, though few users know about this. In any case, it is outside the control of the author. Typically, the font used is some simple sans-serif font in small size, often with a limited character repertoire, especially in older browsers.

This implies that the repertoire of characters you can use there may vary. Other characters may thus be absent, with e.g. small boxes shown instead.

Partly for reasons like this, authors are more and more moving towards using other techni­ques than the title attribute, namely “CSS tooltips” (or maybe “JavaScript tooltips”). This lets you use the same fonts as in the textual content or, if preferred, to set some suitable other fonts.

A simple way to set up a CSS tooltip is to use an attribute of own your, preferably a so-called data-* attribute, say data-tip="...", and use generated content to show its value on mouseover.

Which characters can I use?

There is great variation in font support to characters, from practically universal to prac­ti­cal­ly nonexistent. It is not possible to specify a list of “safe” characters. You need to weigh the importance of using special characters against the risks. Some guidelines and hints: