The key of the degenerate state recognition of roller bearing is feature extraction and pattern recognition. Local characteristic-scale decomposition（ LCD） is a new time-frequency analysis method,which is very suitably applied to the feature extraction of roller bearing vibration signal. Since variable predictive model based class discriminate（ VPMCD） is a pattern recognition method in which the relationship between the feature values is adopted,it can be applied to the he degenerate state recognition of roller bearing. In this paper,LCD,VPMCD and Gaussain mixture model（ GMM） are combined. Furthermore,the degenerate state recognition method of rolling bearings based on LCD and GMM-VPMCD hybrid model is proposed. Firstly,the whole life data of rolling bearing is decomposed by LCD method and the feature values of the components are extracted; then the feature values are clustered in time domain by using GMM and the whole life data is divided into some degenerate states in time domain; finally,the VPMCD model is constructed and applied to the degenerate state recognition of roller bearing. The experiment results show that the GMM-VPMCD hybrid model based on LCD can be effectively applied to the degenerate state recognition of rolling bearing.