Aiming at the problems of poor nonlinear learning ability and low prediction accuracy of shallow prediction models for the remaining useful life of rolling bearings, a method for predicting remaining useful life of servo motor rolling bearings based on stacked gated recurrent unit (SGRU) was proposed. Time-domain and time-frequency domain features were extracted from bearing vibration signals. The common time-domain feature parameters and time-frequency domain feature parameters obtained through ensemble empirical mode decomposition were taken as the original feature set.After that, the similarity measurement method was used to filter the original feature set to get the best feature reflecting the bearing degradation performance.Then, the deep bearing life prediction model was built by stacking two layers of GRU hidden layers, and the network was trained with degradation characteristic parameters as input and the network parameters were optimized continuously. Finally, the proposed method was compared with the LSTM method on the FEMTO dataset.The experimental results show that the proposed method has higher prediction accuracy than the LSTM method.