Late to the question, here, but: I maintain a class at this address: http://lacinato.com/cm/software/emailrelated/emailaddress
It is based on Les Hazlewood's class, but has numerous improvements and fixes a few bugs. Apache license.
I believe it is the most capable email parser in Java, and I have yet to see one more capable in any language, though there may be one out there. It's not a lexer-style parser, but uses some complicated java regex, and thus is not as efficient as it could be, but my company has parsed well over 10 billion real-world addresses with it: it's certainly usable in a high-performance situation. Maybe once a year it'll hit an address that causes a regex stack overflow (appropriately), but these are spam addresses which are hundreds or thousands of characters long with many many quotes and parenthesis and the like.
RFC 2822 and the related specs are really quite permissive in terms of email addresses, so a class like this is overkill for most uses. For example, the following is a legitimate address, according to spec, spaces and all:
" (again)
No mail server would allow that, but this class can parse it (and rewrite it to a usable form).
We found the existing Java email parser options to be insufficiently durable (meaning, all of them could not parse some valid addresses), so we created this class.
The code is well-documented and has a lot of easy-to-change options to allow or disallow certain email forms. It also provides a lot of methods to access certain parts of the address (left-hand side, right-hand side, personal names, comments, etc), to parse/validate mailbox-list headers, to parse/validate the return-path (which is unique among the headers), and so forth.
The code as written has a javamail dependency, but it's easy to remove if you don't want the minor functionality it provides.
