The threshold and confidence of fuzzy production rules are difficult to determined while fault diagnosis of FMS is conducted with fuzzy Petri Nets. A new fault diagnosis method of Self-Learning Fuzzy Petri nets(SFPN) is proposed in this paper. Through slicing processing the model of knowledge, the parameters of weight, threshold and confidence are trained by learning ability of neural network. The process of FMS fault diagnosis based on SFPN is given and system simulation experiments are carried out. Simulation results show the model has self learning abilities of neural network. It can study new expert experience knowledge continuously which breaks the limitations of weight preestablishing in traditional fuzzy Petri Nets. It has certain intelligence.