In order to realize effective fault diagnosis of transformer on-load tap-changer, a fault diagnosis method is proposed, which is based on variational mode decomposition (VMD) and Fisher-Score feature selection. Firstly, the signal is decomposed into a series of modal components with narrow band and distinguishing center frequency by VMD method. Then the energy and singular values of each modal component, and the permutation entropy and singular spectrum entropy of the reconstructed signal are calculated, and Fisher-Score method is used to select the features with better discriminative ability to construct feature vector group. Finally, the feature vector group is input to the support vector machine (SVM) optimized by simulated annealing algorithm to realize the classification, recognition and fault diagnosis of the vibration signals. An experiment is carried out on the transformer on-load tap-changer simulation experiment platform and the collected signals are processed. The results show that the proposed fault diagnosis method has higher identification accuracy.