Current model-based approaches to software debugging use static program analysis to derive a model of the program. In contrast, in the software engineering domain diagnosis approaches are based on analyzing dynamic execution behavior. We present a model-based approach where the program model is derived from dynamic execution behavior, and evaluate its diagnostic performance on the Siemens software benchmark, extended by us to accommodate multiple faults. We show that our approach outperforms other model-based software debugging techniques, which is partly due to the use of De Kleer’s intermittency model to account for the variability of software component behavior.