It is difficult for the empirical wavelet transform (EWT) to extract the fault feature of bearing weak fault in the strong noisy environment. In the light of this problem, a new rolling bearing weak fault diagnosis method based on the probabilistic principal component analysis (PPCA) and EWT is proposed. First, the raw signal is analyzed using PPCA to extract fault feature and restrain noise interference. Second, the signal is decomposed using EWT. The most qualified components are selected to reconstruct signals using the correlation coefficient-kurtosis criteria. Finally, the envelope spectrum is performed to extract fault features of the rolling bearing signals. The simulation and experimental data are analyzed using the proposed PPCA-EWT and EWT-based envelope analysis. The results show that the noise is eliminated and the fault feature is enhanced by the PPCA-EWT process. is the method is effective in the rolling bearing weak fault detection.