I read the s*&t out of them, writing in the margins, highlighting cool quotes from my favorite players, etc. I have a system of symbols/keywords for the main themes of games, and after finishing one write down all of them in the margin next to that game. That way later, if I'm looking for a game exemplifying, say, good kingside attack technique, or a powerful knight outpost, it's easy to find.
To be honest, I like studying chess more than playing it!
I mean, playing is great, but the difference between playing over a classic game and playing one myself is like the difference between farting around on my guitar vs listening to a virtuoso play.

I gotta admit - I don't have many chess books (6) but I actually read 'em, and they help! Freaking time consuming for the game collections (still not done with 'em), but I definitely improve every time I study them.
I've been reading this book on the Modern Defence lately that's definitely over my head, but I'm discovering that a combination of first studying the annotated games and using an engine to 'refute' your plausible alternative moves, then following up with repeated goes on a "guess-the-move" program works extremely well for studying annotated games. The guess-the-move program really helps for staying honest and not getting lazy about predicting the following moves.
I think it's working, because even though I've just started playing the Modern defense for about 2 weeks, I've gone from getting massacred EVERY game, to beating solid opposition my level in good complicated modern type positions and having as good winning chances as my white repertoire.