The early weak fault recognition of the rolling bearing under low signal to noise ratio is always a difficulty problem. The envelope spectrum method based on hilbert transform demodulation is a classical bearing fault detecting method which is applied on engineering practice widely. However, the method is hard to deal with the early weak fault diagnosis of the bearing. Thus a high-order statistic called spectral kurtosis is used to study the early weak fault recognition of the rolling bearing. The whole life cycle data from rolling bearing run-to-failure test is used to analysis. The result shows that the spectral kurtosis method can recognize the early weak fault successfully, and detect the fault 200 minutes in advance comparing to the envelope spectrum method, due to its advantage of recognizing the weak signal which is located in the resonance band with high signal to noise ratio.