A common simplifying assumption that there exists, at most, a single fault in the system at any given time does not hold for complex systems with large number of components or systems with little opportunity for maintenance during operation. By employing concepts from rough sets theory, information theory and heuristic search approach, a novel sequential testing algorithm based on rough sets is presented for multiple fault diagnosis, which can efficiently improve the diagnostic precision, substantially reduce the expected testing cost and realize the real-time dynamic diagnosis independent of the priori probability of system.