In this paper, a method based on LTSA and VMD is proposed to effectively extract fault feature information of rolling bearing. The selection of reduced dimension d and neighborhood selection parameter k in LTSA adopts grid search method. Taking the fault characteristic energy ratio (FER) as the objective function, a set of optimal parameters is determined as the input parameter of the manifold learning algorithm. Variational modal decomposition (VMD) is used to denoise signal. The kurtosis value, correlation value and envelope entropy value are comprehensively evaluated, and the optimal component is selected for reconstruction. The optimal component of reconstruction is analyzed by envelope spectrum to realize fault diagnosis. The proposed method is applied to the fault measurement signal. The results show that the proposed LTSA-VMD fault diagnosis method has obvious advantages in signal denoising and fault feature extraction.