The bearing vibration of wind turbines is nonlinear and non-stationary. To effectively extract bearing vibration signal features for fault diagnosis, a method of feature vector extraction based on variational mode decomposition (VMD) and energy entropy is proposed. In addition, the support vector machine (SVM) classifier is used to identify the types of vibration faults. VMD transforms the constrained variational objective function into the unconstrained one to optimize solution. Compared with the VMD and empirical mode decomposition (EMD), the modal decomposition layer of VMD is less than EMD, no found false modality, and truly reflecting signal components. After the Hilbert transformation, double logarithmic coordinates show that the VMD-based spectral characteristics are significant. VMD is performed on the different types of vibration signals of wind turbines. Therefore, VMD is not that affected by noise and has few decomposition levels. The energy entropy of the normalized four modal components is considered the eigenvalue and classified by SVM, and compared with EMD-based and wavelet db4-based energy entropy eigenvalue extraction methods. Experimental results indicate that the accuracy of the method is higher than those based on EMD and wavelet db4, under the limited sample condition. Thus, a referential diagnostic method is provided for practical applications. (C) 2019 Elsevier Ltd. All rights reserved.