Aiming at the traditional artificial neural network (ANN) has the problem of easily falling into the local minimum and the slow convergence, a fault diagnosis model based on Tabu search algorism combined with ANN was proposed. Firstly, the fault diagnosis model was defined, then the trained data was used to train the weight and the threshold of the network, and the trained result was used as the initial solution of the improved tabu search algorism, the definition and description of the algorism were given. Finally, the global optimization result was set to the neural network, and the test data was used to as the input of the neural network to diagnose. The simulation result shows that our method in this paper conquers the defects of the traditional methods, has the advantages of high diagnosis accuracy, rapid diagnosis speed and fast convergence.