The invention provides a modular BP neural network fault diagnosis method based on fault propagation, belonging to the technical field of fault diagnosis and positioning of an electronic system. Based on circuit module division, the test node of each sub circuit is determined, and characteristic parameter data of the test nodes is taken as a data source of the fault diagnosis; a modular abnormality detection model is established based on the data source of circuit simulation, the fault propagation is analyzed, and a modular BP neural network model is established; when an actual circuit is failed, the modular abnormality detection model is used to carry out first level positioning, a failed sub circuit is determined, then a BP neural network model of a target sub circuit is used to carry out second level positioning, and a fault mode is identified. According to the method, the fault positioning is carried out through a model established in an offline way, the expansibility is strong, the application range is wide, the real-time fault diagnosis of a large-scale digital and analog mixed circuit in an online way is realized, and the method has very high accuracy of fault positioning especially for a condition of fault propagation.