Rolling bearing is the core component of large rotating machinery, and the fault diagnosis of rolling bearings is of great significance to guarantee the stability of rotating machinery. EMD is a very effective method for fault diagnosis because of its unique advantages in the analysis of nonlinear unstable signal. However, the in- trinsic endpoint effects of EMD cause non-negligible feature extraction errors, which will affect the accuracy of fault diagnosis. To solve the above problem, this paper proposed an EMD method based on undistorted source signal whose endpoints are extreme points(UEE-EMD). EMD restrained the emergence of endpoint effects from the source through overlapping sampling and extremal endpoints, obstructed the signal distortion by cutting the ends of IMFs, and guaranteed the accuracy of the feature extraction. The simulation experiment of fault diagnosis shows that the rolling bearing fault diagnosis based on UEE-EMD achieves better diagnosis results.