Abstract: First, a method to construct hierarchical models of a system for fault detection and diagnosis (FDD) is presented, based on consideration of the function feature of faults and accurate definition of the conception of mode. Second, a hierarchical FDD strategy is also presented, which uses the constructed models as fundamental descriptions of the system, and de kleer's general diagnosis engine (GDE) as the fundamental FDD tool. The hierarchical scheme can greatly improve diagnosis efficiency. Accurate definition of mode makes the system's hierarchies clear, prevents the algorithm from redundancy and produces compact probability transmission formulas.