EMD is now a commonly used nonlinear and instable signal processing method, but it has boundary runaway and modal aliasing. The single disunited IMF cannot well reflect the characteristics of the respective vibration source. Therefore, in order to suppress the boundary runaway that will appear in the process of EMD, the image method is used to extend the length of signal data. To solve the modal aliasing, it is necessary to decompose the extended data by the EMD method, to distinguish the IMF that produces modal aliasing after decomposition, to integrate it according to the integrity of the EMD and then to re-decompose it after adding broadband white noise with the average value of zero. On the basis of that, it is better to improve NS-EMD method and realize the AM-FM demodulation by standardized method. By the spectrum analysis, we extract the fault characteristics of rolling bearings and propose a method to diagnose faults of rolling bearing. The results of analyzing the simulation and the vibration signal of fault rolling bearing shows that the method can effectively extract fault characteristics of rolling bearing.