Abstract: This work aims to puts forward a reliability assessment model based on Fuzzy Neura1 Network(FNN), and to present a modified training algorithm. The modified algorithm combines supervised learning scheme with self-organized learning scheme, which feedback errors and modify parameters, at the same time, automatic adjust the central values and the width of membership function. Finally, the results of simulation show that the proposed model has the advantage of faster learning rate and higher assessing precision over BP model and SOFM model.