This is a wonderful little gem of a book that presents the theory of computation in a fascinating way. It is targeted at advanced undergraduates in computer science, but assumes remarkably little prior knowledge, making it accessible to nearly anyone. The book covers a lot of ground, including the standard fare of automata, computability, and complexity results, plus some bonus material such as probablistic and parallel complexity, information theory, decidable logical theories, and other topics that are normally left out of introductory books. On top of this, the book is remarkably thin!
The best attribute of Sipser's book, though, is the engaging style. This is an easy book to read. You will not feel like you're running into a brick wall, as is sometimes the case with books on abstract topics. It's not so much that the book is slow or gentle (it's really not) as that it is interesting, engaging, and has a knack for stopping short of getting too caught up in details. A number of small things -- the occasional amusing exercise, the "proof idea" sections, or helpful pictures -- add up to an enjoyable reading experience.
Two cautions are appropriate to students considering this book. First, there are variations between authors in the definitions of various automata (especially PDAs). The differences are trivial, and more a matter of taste than of any real importance; but it could come up if you use Sipser as a supplement to a course that follows a different textbook. Second, the coverage of many topics in Sipser's book is brief and concise, sometimes more than you might like. Some important concepts (for example, pairwise distinguishability of strings) are only mentioned in exercises, not in the main chapter, so at least skim all the exercises even if you don't do them. The sketchy coverage is especially pronounced in advanced topics, so (as always) expect to do some filling in of concepts if you go on into further study of this area.