It's very hard to choose just one. If I had to choose just one it would be:
The Book of the Righteous, Green Ronin, written by Aaron Leob. This is the single best rules supplement ever issued for an RPG. It could readily be used as is, but the best thing about it is the overall approach he lays down and the maturity and seriousness with which he addresses the subject. It's a shame Aaron didn't write more.
After that, it's a long list, and I'll have to stick to D&D or I'll end up listing 50 books.
1e AD&D DMG, TSR, by Gary Gygax. This is the obvious choice. This is such a good read filled with so much creativity and covering so much breadth. This book still manages to set a standard that other systems struggle to reach. Would that we'd see more small dense black type text written by someone that can actually write rather than wasted money on all this gloss and $50 price tags for things with 1/10th the crunch.
Beastiary, The Predators, BetaBunny Publishing: Every monster manual should be this awesome and this crunchy. Rivals the 1e MM, the 2nd edition Monstrous Compendium, and the Tome of Horrors but in some ways goes one better than each of them. Much like with Aaron Leob, I regret that the author didn't put out more. One of the few pdfs I've never regretted buying. Heck, in fairness I ought to tip the author another $10 - and I would if I could get refunds for some of the junk I spent money on.
1e AD&D MM
The Complete Shaman's Handbook, Green Ronin, by Steve Kenson - Excellent material, excellent reuse of material in an unique way. One of the few books that was adopted into my house rules with almost no ammendments. If you know how picky I am about rules and how arrogant I am, then I hope you understand how high praise that is.
3e Player's Handbook, by Monte Cook - Revitilized and revolutionized not only the system but the hobby.
Hot Pursuit, The Definitive D20 Guide to Chases - There are a few small flaws in the rules as presented, but no other book has ever provoked so much thought from me about how to build systems and why. The general concept is as solid as it simple and I think it ultimately over turns the implicit or explicit notion that has bedevilled designers for decades that the best system is a system that handles everything through a centralized generic mechanic. Instead of trying to shove every problem in the same round hole, I realized that it was a far better design to have in a sense a bunch of minigames that were specialized to represent the particular scenario you were trying to address. The idea of one set of rules for combat, mass combat, general propositions, social challenges, chases, domain governance, etc seemed so old fashioned. Of course you needed different rules, because the central idea of the contest differed.