The traditional BP neural network classification algorithm has the inherent shortcomings such as miscarriage of justice when each network's output value or network output value between the mode classes is similar. So the fault diagnosis method of BP neural network based on mode correlation is proposed. Then the modeling and simulation are performed with a specific circuit. Experimental results show that the BP neural network which combining with mode correlation has made full use of all of the value which is outputted by every node of the network output layer and strengthened the network output performance. This method has made mode classification more convenient, better classification results and has certain universality.