HTML5: Up and Running is the book version of Mark Pilgrim’s comprehensive introduction to HTML5 at DiveIntoHTML5.org. Whether you buy the book or read it online, it’s the best introduction to the topic you’ll find.
Mark begins with the history of HTML5 (using email archaeology, as he calls it). You’d never guess that many of the problems we have with XHTML, MIME types, etc. have roots in discussions over 20 years ago. From then on, he moves into feature detection (which uses the Modernizr library), new tags, canvas, video, geo-location, storage, offline web apps, new form features and microdata. Each chapter can be read independently – so if you’re planning to use this as a reference, you may be better of reading the links kept up-to-date at DiveIntoHTML5.org. If you’re interesting in learning about the features, it’s a very readable book, terse, simple, and above all, delightfully intelligent.
Incidentally, if you’re starting off on a new HTML5 project, you’re probably best off using HTML5BoilerPlate.com. It’s very actively maintained, and contains some really nifty tricks you can learn like the protocol-relative URL.
Disclosure: I’m writing this post as part of O’Reilly’s blogger review program. While I’m not getting paid to review books, I sure am getting to read them for free.