Aiming at the problems of early bearing fault signal nonlinearity, nonstationary and difficult to extract fault features, a feature extraction method combining variational mode decomposition (VMD) and manifold learning was proposed. In this method, VMD was used to decompose signals into intrinsic modal components containing different fault information, then the features were extracted from them and a highdimensional mixed domain feature set was constructed. Finally, the optimized manifold learning isometric mapping algorithm (ASL-Isomap) was used to reduce the highdimensional feature set into a lowdimensional mixeddomain feature set with better fault discrimination, and the support vector machine (SVM) was used to implement fault classification and recognition. The rolling bearing experimental results show that this method can be used to accurately and clearly extract fault feature information, and the diagnostic accuracy is higher than that of traditional methods.