This kept getting recommended in random places, so I thought I'd get it (Tufte is also famous for 'The Cognitive Style of Powerpoint'; he seems to hold it with the same regard Dijkstra had for COBOL). I wasn't quite sure what to expect, but it wasn't quite this. It's wonderfully laid out, trying to practise what it preaches in terms of clear communication. Ironically, it strongly emphasises that graphs should make the reader think about the data, not about the layout, in a book full of graphs where you'll be studying them for the layout. So, the most distracting examples are the best, since you end up thinking about the data rather than the presenation.
This is not a book of hard and fast rules. It seems more about encouraging people to think. Tufte is very big on x-y plots which can show interactions between variables far better than a pile of bar charts, and is very much against 'chartjunk'. That's not terribly surprising, but the way he presents it works well. Some bits are strangely personal. For example, he complains of the redundancy in a standard box plot, and pares it down to a few line segments and dot, apparently missing that the box plot could be viewed as a very stylised presentation of a density function, so that his rearrangements lose a great deal of the intuition the presentation provides.
Strangely enough, the book this reminds me of the most is Raskin's 'The Humane Interface'. It's a mixture of rallying against the low standards commonly accepted, talking about non-controversial improvements, and proposing a slightly more extreme or eccentric personal variation. Also, there's an element of Knuth to it, but what do you expect given that Tufte remortgaged his house to fund the self-publishing of the book to the standards he wanted. That's a man who cares about presentation.