Fault diagnostic system is of great importance in monitoring and controllingspacecraft in the ground control center. The bottleneck problem of knowledge acquisition for spacecraft fault diagnosis is solved by using fault tree knowledge. Thepaper presents a fault diagnostic method based on fault tree and neural networkmodel. Based on the hierarchical model of fault tree, knowledge representationmethod based on frame and generalized rule is presented, and the relevant certainand possible reasoning strategies are described. Learning diagnosis based on neuralnetwork model is used to confirm and verify the results from the possible reasoning. By using Borland C++ under Windows, a fault diagnostic prototype systemis developed, and the validity is also demonstrated by diagnosing a satellite powersystem fault imitation bench.