A fault diagnosis method for wind turbines gearbox bearings was proposed based on information fusion combining BP neural network and D-S evidence theory. Firstly, based on big data, the fault characteristics of vibration, temperature, current, torque and rotating speed signals related to the faults of wind turbines gearbox bearings in SCADA system were explored. Then, the fault feature quantity of each signal was used as the input of the neural network, and the outputs of the neural network were normalized as the Basic Probability Assignment (BPA value) of D-S evidence theory. In order to solve the conflict between evidences, a weighted-based improvement method was used to improve the evidence. Finally, the combination rules were used to fuse the improved evidences to obtain final diagnosis results. The study was based on actual operating data of a 2 MW wind turbine in a wind farm, and the results show that as the dimension of the fusion signal increases, the accuracy of the final diagnosis will gradually increase. The reliability of fusing multi-dimensional signals is significantly higher than that of single signals.