Aiming at the disadvantage that the extraction features may contain smaller correlation and redundancy characteristics in the rolling bearing fault diagnosis, distance evaluation technique (DET) based on Wrapper was used for feature selection. In the design of the classifier, regression-Variable predictive mode based class discriminate (RRVPMCD) was put forward to reduce the effect of abnormal value in the estimation of parameters, therefore more accurate prediction models can be built up. According to the characteristics of the Wrapper mode, firstly, the sensitivities of each feature was calculated for class by using DET，and several features with the biggest sensitivity were chosen to establish feature vector matrix combined with RRVPMCD，then, a predictive model was built with the method of RRVPMCD, finally, the established predictive model was used for pattern recognition. Experimental results show that the model based on the Wrapper feature selection and RRVPMCD method can effectively identify work status and fault type of rolling bearing.