We present an Artificial Intelligence based method for improving the reliability of software applications, especially in digital cable TV set-top-box and other embedded environments. Initially a small finite state model of the software system and all relevant applications is constructed to define all user input events and application states of interest. A small set of expert system rules is then defined that analyzes state transitions in testing data. When these rules are applied to actual testing data a quantitative measure of suspicion is assigned to all event transitions in the original finite state model. Analysis of this annotated model can then uncover the source of otherwise intermittent inter-application failures.