When the existing fault diagnosis methods are applied to an aircraft with a large number of components associated with each other, there appear space explosion problems in those diagnosis methods based on fault tree analysis (FTA), and the difficulty in sorting the training samples in methods based on neural network. This paper proposed a scheme that integrates the fault tree with BAM neural networks, in which all the failure modes of a system are summarized with the fault tree, sorting out the necessary training samples for the BAM neural networks. On this basis, fast and accurate aircraft fault diagnosis can be achieved by applying BAM neural networks. The experiment evaluations show that the proposed method has better scalability, and the average fault-judging rate is improved by 20%.