Considering training time of traditional BP neural network is too long and it cannot solve the problems in the input vector with multiple-valued, a new method of BP neural network based on rough neuron is presented. A rough neuron can be viewed as a pair of neurons. One neuron corresponds to the upper boundary and the other corresponds to the lower boundary. Upper and lower neuron exchange information with each other during the calculation of their outputs. Firstly, the continuous attributes in diagnostic decision system are discretized with particle swarm optimization. Then, the reducts are found based on attribute dependence of rough set, and the optimal diagnostic decision is determined. Lastly, according to the optimal decision system, rough neuron network is designed for fault diagnosis. A practical example is given , the method is feasible and available.