I'm going to express size in Knuths, where one Knuth is the size of one volume of Donald Knuth's "The Art of Computer Programming."
Knuth's work is THE seminal text for this sort of thing. It's very long -just over 4 Knuths thus far, and Volume 4 is actually only the first subvolume of something much larger- and it's very, very dense. If you want The Big Book of Algorithms, it certainly fits the bill, but does so in the almost cartoonish sense of the massive book that breaks whatever desk you put it on. It may also break the bank: this many books of this quality and length can get expensive.
Introduction to Algorithms, by Cormen, Leiserson, Rivest, and (in later editions) Stein, also has a very good reputation. It's about 1.5 Knuths long, and it's not as dense, so you're not going to find as much raw STUFF in here, but it covers the basics quite well, and also a fair amount of the not-so-basics.
O'Reilly's "Mastering Algorithms with _____" (there are versions for a number of languages) can also be good for quick reference. Length-wise, they're about a Knuth apiece, and there's only so much you can fit in a book that size. On the other hand, unlike the previous two that I've mentioned, these books present algorithms in a language people actually use. CLRS uses pseudocode, while Knuth himself often uses a kind of assembly language (I told you it was dense).
Speaking of O'Reilly, they also have "Algorithms in a Nutshell." This is the shortest one I can really talk about here -about 0.5 Knuths, maybe shorter- and that really limits what you can put in it, but it hits the most common cases.
