A method for rolling bearing fault auto-diagnosis based on wavelet packet coefficient, kurtosis and envelope analysis is presented according to rolling bearing vibration feature. First, the fault signal will be decomposed into different frequency bands by using wavelet packet. Then, the kurtosis of different frequency band is calculated. In order to avoid envelop spectrum analyses for all wavelet packet coefficient at different frequency band, a criteria in terms of the maximum of kurtosis factor (CMK) is used for selecting the frequency band for analysis. The one containing abundant fault information is chosen automatically by CMK. In the end, the fault frequency can be determined by the envelop spectrum of the chosen group of wavelet packet coefficient. This method can improve reliability and is convenient for rolling bearing fault diagnosis.