Over thirty years of research have established {\em program analysis} as the `art' of extracting run-time characteristics of a program already at compile time. The wide spectrum of programming languages and paradigms led to a large body of theory, algorithms and implementations, each typically developed for a specific scenario. Only recently the investigation of this body with respect to abstract {\em profiles} has begun, with the goal to support cross-fertilization between the involved communities. This investigation does not only justifiy the development of the various different methods and algorithms. It also reveals the power of {\em heterogeneous} approaches to program analysis that make use of a number of different analysis methods. The talk will focus on the necessity of profile-based, heterogeneous program analysis methods, and their impact {\em beyond} their original application domain. In particular, it will be shown that a careful balance of generality and precision may overcome some at first sight unavoidable hurdles, concerning both computability and practical acceptance.