Model-based diagnosis is an active field of artificial intelligence research. Diagnostic problem is characterized by a set of observations to account for the differences between the system’s actual behaviors and its expected behaviors. The classical method was built on the well-known consistency-based theory and it described the system’s structure and behaviors usually in the first-order language. But it is not efficient and therefore is difficult to be applied in actual systems. Different from the classical method, diagnosing with value propagation is an effective procedure-oriented method. For some special systems, the algorithm terminates in polynomial time. But it is not complete and the system model can only discribles normal system behavior. This paper presents an expanded system model based on value propagation, which can discrible multi-fault mode of the system behavior, redefines the system diagnosis and discusses the relation between consistency-based (abductive) diagnosis and value propagation-based diagnosis. Furthermore, the relation between value propagation and the actual diagnosis has been analysed, which is useful in diagnosis test. Finally, the sufficient condition of the algorithm’s completeness is given, which is valuable for promoting the application of value propagation-based diagnosis.