We propose a slip vector construction method for fault diagnosis that is based on singular value decomposition theory and decomposition matrix frame analysis. Per this method’s guidelines, we introduced the main singular value ratio, maximum eigenvalue reconstruction and optimized filter design methods. We successfully applied the proposed method to the fault feature extraction of rolling bearings. The experimental data analysis results showed that this method has suitably able to extract weak shock fault features. This paper has important implications in intelligent fault diagnosis of rolling bearings in circumstances of strong noise.