This is hands-down the best book on statistics I have ever come across. It doesn't get bogged down in unnecessary mathematical detail, nor does it patronise the reader with trivial examples. Somehow the author manages to communicate concepts intuitively and efficiently, without seeming dry. If you are looking for a swift and clear way to learn statistics, this is the book for you.
Disclaimer: this book is aimed at advanced undergraduates and beginning graduate students who are looking to learn some statistics for application in computer science. If you are a hard-core mathematician, you are likely to find it frustratingly non-rigorous. Likewise, if you are an untrained scientist, you may find the mathematical style alien. But if, like me, you are a theoretical physicist, the material is refreshingly light, the approach is pleasingly logical, and unnecessary calculations are left to the reader - a perfect balance for serious study.
Looking at the table of contents one would think that this is the one book one would need to become a statistician. The subject is built up from elementary probability theory, and the book goes all the way to Monte Carlo Markov Chain methods.
The author manages to cover a lot of statistical theory in 442 pages. He does this by giving most theorems without proof, but often with a rationale for having the theorem. One can see where the journey is going, but the reader will have to take many of the actual steps for him- or herself. In the exercises the reader is asked to supply some of the missing proofs, and it is often a good idea to try and prove a theorem even if one is not explicitly asked to.
Its conciseness makes the book useful as a reference manual for people who already know statistics. Learning statistics from it requires a fairly solid mathematical background and a lot of effort.