I’ve tested all exploits I know on HTML Purifier and it did very well. It filters not only HTML, but also CSS and URLs.
Once you narrow elements and attributes to innocent ones, the pitfalls are in attribute content – javascript: pseudo-URLs (IE allows tab characters in protocol name – java script: still works) and CSS properties that trigger JS.
Parsing of URLs may be tricky, e.g. these are valid: http://spoof.com:xxx@evil.com or //evil.com. Internationalized domains (IDN) can be written in two ways – Unicode and punycode.
Go with HTML Purifier – it has most of these worked out. If you just want to fix broken HTML, then use HTML Tidy (it’s available as PHP extension).
