A new rolling bearing fault diagnosis model is proposed based on partly ensemble local characteristic-scale decomposition (PELCD), Laplacian score (LS) for feature selection and variable predictive model-based class discrimination (VPMCD). PELCD overcame the mode mixing of LCD and excelled the traditional ensemble noise-assisted method. In this paper it is employed to deal with the rolling bearing vibration signal. Then the features in time and frequency domains of comments are extracted. Besides, LS is utilized for feature selection to decrease the dimension of feature vector and improve the efficiency of fault diagnosis. The selected features are inputted to the VPMCD-based classifier for training and testing. Finally, the proposed model for rolling bearing fault diagnosis is applied to experimental data and the analysis results indicate that the proposed method can identify fault categories and degree effectively.