ATo reslove the recognition problem for different fault types of roller bearings, this paper proposed a method based on Variational Mode Decomposition (VMD), Base-scale Entropy (BSE), and GG clustering algorithm for roller bearings diagnosis recognition. Firstly, the method used the VMD model decomposed the roller bearings vibration signals in different conditions into a series of intrinsic modal functions (IMFs). Secondly, determine the number of decomposition layers based on the center frequency, and it used BSE method to calculate the entropy value of the first three IMF to construct feature vectors [BSE1, BSE2, BSE3]. Then GG clustering algorithm selected the value of [BSE1, BSE2, BSE3] as the input eigenvectors for fulfill the roller bearings diagnosis recognition. Finally, the results show that combining VMD and BSE can effectively extract feature vectors of different fault types and the fault recognition for roller bearings is good by using GG clustering.