While the intuition underlying a zero knowledge proof system [GMR85] is that no "knowledge" is leaked by the prover to the verifier, researchers are just beginning to analyze such proof systems in terms of formal notions of knowledge. In this paper, we show how interactive proof systems motivate a new notion of practical knowledge, and we capture the definition of an interactive proof system in terms of practical knowledge. Using this notion of knowledge, we formally capture and prove the intuition that the prover does not leak any knowledge of any fact (other than the fact being proven.) during a zero knowledge proof. We extend this result to show that the prover does not leak any knowledge of how to compute any information (such as the factorization of a number) during a zero knowledge proof. Finally, we define the notion of a weak interactive proof in which the prover is limited to probabilistic, polynomial-time computations, and we prove analogous security results for such proof systems. We show that, in a precise sense, any nontrivial weak interactive proof must be a proof about the prover's knowledge, and show that, under natural conditions, the notions of interactive proofs of knowledge defined in [TW87] and [FFS87] are instances of weak interactive proofs. © 1988 ACM.