When the rolling bearing works in fault condition, the complexity of the vibration signal will change. Sample entropy (SE) is defined to measure the complexity of time series in single scale, while multiscale entropy (MSE) is used to measure the complexity of time series in different scales, which contains much more information. Based on this, a new rolling bearing fault diagnosis approach based on MSE and SVM was put forward. In this paper, the concepts of SE and MSE were introduced, then MSE was applied to extract the feature information from bearing vibration signals and SVM was used to identify the rolling bearing fault categories. Finally the proposed approach was applied to the experimental data, and the analysis results indicate that the proposed approach can fulfill the rolling bearing fault classification effectively.