In order to enhance life prediction accuracy of rolling bearing， as well as its degradation performance trend and fluctua? tion range， a remaining useful life prediction method of rolling bearing based on LSTM-ES-RVM networks is proposed. The LSTM model is used to learn frequency domain and establish the preliminary Health Indicator （HI）， so as to reduce dependence on prior knowledge and artificial experience. An improved local extremum model is proposed to eliminate violent oscillation and im? prove the global monotony of Health Indicator. The RVM model is used to predict remaining useful life of rolling bearing. The ex? perimental results show that the proposed method achieves the optimal prediction performance compared to other prediction meth? ods in the field.