Abstract The prediction accuracy of bearing remaining useful life (RUL) is not high due to the unreasonable stage division of bearing performance degradation and the blindness of feature selection. In order to solve this problem, RUL prediction for rolling bearings using Pearson product-moment correlation coefficient (PPMCC) and Kullback–Leibler divergence (KLIC) feature selection is proposed in this paper. First, in order to divide the bearing performance degradation status more accurately, a novel performance degradation state partition method is provided based on t -distributed stochastic neighbor embedding and a k-means clustering algorithm. Second, multi-scale time status feature selection is implemented using PPMCC and KLIC. Finally, the above-proposed feature selection method is validated on the performance degradation data of rolling bearings provided by the University of Cincinnati platform. The results show good and stable effectiveness of the proposed method, and its superiority is demonstrated by comparison with other approaches.