I'd heard quite a few good things about this book, so I decided to buy it. It's very informally written, but has (I think) a pretty strong underlying argument.
Koster's interested in making computer games more fun. He takes the perspective that what is enjoyable when playing is the learning aspect of the games. Using this as a starting point, he considers ways that games succeed and fail to be fun. He then goes on to think about how games might be improved in the future.
After just one read, I don't really feel ready to comment on his thesis. I'll need to go over it once or twice more to get my head around the subtleties, I suspect. It's a pretty short book, though, so that's not really going to be a problem.
That said, he definitely raises some very interesting points. Also a few things that are applicable to tabletop roleplaying games as well.
A Theory Of Fun (Amazon)