For vibration signal generated by rubbing fault in rotary machinery including the single-point rubbing and the partial rubbing, multi-component harmonic frequencies are presented. According to the vibration response of the rotor rub-impact, wavelet packet energy method is used to extract the frequency character from rub-impact fault signal. It can resolve the difficulty of mul-ti-component fault feature extraction of rubbing fault and multi-faults with rub-impact. In this paper, the rubbing fault signals collected by displacement sensor and acceleration sensor were decomposed clearly into different frequency band by using wavelet packet method. The vibration signals from experiment rig of rub-impact rotor include just rub-impact and multi-faults with imbalance and rub-impact. The energy strength belonged to different fault frequency in the frequency band is obtained respectively. The analysis results of the rubbing signals indicate that the multiple features can be better extracted with the wavelet packet energy, and it is effective in analysis and recognition of rubbing fault.