In the paper, the scheme of establishing a self diagnosis model for high speed rolling bearing is proposed considering the complexity of rolling bearing fault diagnosis and combining with the difficulties of rolling bearing fault diagnosis for equipment on site. The original vibration signal of rolling bearing for high speed gear box is decomposed using the wavelet packet, then frequency domain signals in different frequency bands in decomposed original vibration signal are reconstructed and the characteristics of shock pulse energy for rolling bearing fault of reconstructed frequency domain signals are extracted respectively. The extraction of characteristics for rolling bearing fault of high speed gear box is realized and they are used for establishing the self diagnosis model for high speed rolling bearing through comparing the characteristics of impact energy for rolling bearing extracted from each frequency band after reconstruction as well as extracting frequency band signals sensitive to energy changes aiming at different types of rolling bearings.