In view of limitations of the circuit board fault diagnosis technology on infrared images, in this paper, the intelligent diagnosis method is analyzed. In the method of neural networks, the multiple classifiers are turned into a dichotomous thinking, and an integrated neural network diagnosis model is designed based on BP neural network. For the same type of faults, samples within a range are trained in the network, and for each group of the measured fault data and the several sub-threshold selected, the diagnosis is made according to the characteristics. Finally, the living examples are simulated and tested by using MATLAB. The results show that the recognition accuracy is improved，the detection sensitivity can be increased by 1.74 times, and the prediction error is decreased to 17.6 % of the original prediction error of the more-fault-mode network. This provides a theoretical basis for the practical circuit fault diagnosis.