Abstract: To solve the defects of traditional fault diagnosis neural network, such as long training time, complex structure and single-valued input, a fault diagnosis system for civil aircraft based on rough-neural network was proposed. Rough set theory was applied to the front-end neural network to reduce the data of civil aircraft fault sample so as to remove the disturbance of redundant attributes, and overcome the impaction of unrelated data that imposed on the performance of network learning, simplify network structure. Secondly, by using the rough neurons instead of the traditional neurons, the performance of network was improved, and the scope of the application of network was expanded. The effectiveness of this method was verified by Airbus A320 aircraft fault diagnosis test .