When a rolling bearing fault signal is weak and masked by background noise will fail to fault diagnosis. This paper presents a novel method to filter the fault signal to enhance the fault features based on robust Principal component analysis (RPCA). RPCA is designed for addressing the problem that signals have outliers, and it decomposes the signal to a low-rank matrix and a sparse matrix. If the impulse signals are treated as the outliers, the RPCA operations can extract the fault signal component. The extracted signal is performed to reconstruct the signal can obtain the fault signal, which can be used for diagnosing successfully. The effectiveness of the proposed method is validated by outer race defect, inner race defect, and roller defect fault signal. The results show that the filtered signal envelope spectrum can accurately find out the fault characteristic frequency of bearing fault, which verified that the proposed method has an excellent performance to diagnose the bearing fault.