The Drunkard's Walk
Mlodinow gives lots of examples of how people have got it wrong when dealing with probabilities. You may have come across some of them before - the Monty Hall problem is a classic one - but you're likely to find some which you haven't seen before. I don't think that he get's everything right though. He claims that a lottery in 1920's New York which used the last 5 digits of the US treasury balance could have been gamed by someone who knew Benford's law - sorry I don't buy that.
The book goes on to describe some of the history of probability and statistics - the lives of Cardano, Pascal, and the Reverend Thomas Bayes. Mlodinow also looks at how statistics became important in running our lives, and at what can go wrong when we try to measure something.
I would say that the final chapter is the most important one of the book. Here Mlodinow argues that understanding randomness isn't just about obscure puzzles - its central to what happens in our lives. Analysis made with hindsight may just be trying to make sense of random events - but people believe it, and it may well become self fulfilling. It's an important message, in a book which is enjoyable and easy to read.
Note: This book is on the shortlist for the 2009 Royal Society Prize for Science Books