Preventing improper information leaks is a greatest challenge of the modern society. In this paper we present a technique for measuring the ability of several families of adversaries to set up a covert channel. Our approach relies on a noninterference based formulation of security which can be naturally expressed by semantic models of the program execution. In our analysis the most powerful adversary is measured via a notion of approximate process equivalence. Even if finding the most powerful adversary is in general impractical, we show that this requires only a finite number of checks for a particular family of adversaries which are related to a probabilistic information flow property. Keywords Covert Channels