In order to improve the accuracy of motor bearing fault diagnosis, and aiming at the problem of unstable vibration signals and the difficulty in extracting fault feature of motor bearing fault, a motor bearing fault diagnosis method based on the combination of variational mode decomposition (VMD) energy entropy and convolutional neural network (CNN) is proposed. In order to reflect the characteristics of faults more accurately, a three-dimensional energy entropy extraction method is adopted to divide the bearing faults into three categories, namely, inner ring wear, outer ring wear and cage fracture. Then each category is subdirided into three cases with loads of 0%, 25% and 50%, for a total of 9 cases. Firstly, the VMD method is used to decompose the fault signal into components of the intrinsic mode function (IMF) and the energy entropy of each dimension IMF is extracted to form the feature vector. The results show that the method can effectively improve the accuracy of fault diagnosis.