The invention discloses an avionics system fault diagnosis method based on combination of three methods which are FTA (fault tree analysis), a BAM (bidirectional associative memory) neural network and a BP (back-propagation) neural network. According to the invention, a fault diagnosis model integrating FTA and the BAM neural network and a fault diagnosis model combining the BP neural network are established respectively. When a fault phenomenon is obvious, the fault diagnosis model integrating FTA and the BAM neural network is adopted. Firstly, a fault mode of the system is acquired by using FTA, a training sample of BAM is analyzed and summarized, and finally, BAM carries out parallel association through an associative memory matrix so as to acquire a diagnosis result. When a fault phenomenon is not obvious or a fault example base does not have such a fault, the fault diagnosis model combining the BP neural network is adopted. Learning is carried out on a training sample, that is, data such as voltage, temperature and the like, acquired by the BP neural network, and fault diagnosis is carried out by using the trained network in the end. The two fault diagnosis models have complementary advantages, and make up deficiencies of each other, thereby effectively solving problems existing in avionics system fault diagnosis, and improving the accuracy and the efficiency of fault diagnosis.