A rolling bearing remaining useful life prognostics method based on improved Continuous hidden semi-Markov model（ CHSMM） is proposed,aiming at the problem that the CHSMM algorithm prognostics accuracy is low for remaining useful life of rolling bearings. The feature vectors of the time and time frequency domain are extracted from the vibration signal of bearing and then the PCA algorithm is used to reduce the dimension of the feature vectors. Then,the degradation state recognition model and the remaining useful life prediction model are established based on improved CHSMM into which the Gauss mixture probability density function is introduced aiming at solving the low accuracy of remaining useful life prediction caused by the dwell time probability density function which does not conform to reality. Finally,the whole life cycle data of the bearing is input into the model,and the degenerate state and residual life of the bearing are obtained. The experimental results show that the proposed method can accurately predict the remaining useful life of bearings.Compared with the original CHSMM algorithm,the accuracy of the degradation state recognition is increased by12%,and the accuracy of remaining useful life prediction is increased by 23%.