In this paper, we develop a four-phase model for evaluating architectures for clinical decision support that focuses on: defining a set of desirable features for a decision support architecture; building a proof-of-concept prototype; demonstrating that the architecture is useful by showing that it can be integrated with existing decision support systems and comparing its coverage to that of other architectures. We apply this framework to several well-known decision support architectures, including Arden Syntax, GLIF, SEBASTIAN, and SAGE.