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. An ensemble empirical mode decomposition (EEMD) method was introduced to resolve the difficulty of multicomponent fault feature extraction of rubbing fault. The rubbing fault signals created by simulation were decomposed clearly into different single frequency data by using EEMD method. The vibration signals from experiment rig of rubimpact rotor were then analyzed under different directions rubbing, which include in horizontal, in vertical, in horizontal and vertical at the same time. The vibration strength belonged to different fault frequency in the frequency spectrum was obtained from the intrinsic mode function (IMF) based on EEMD respectively. The analysis results of the rubbing signals indicate that the multiple features can be better extracted with the EEMD, and is effective in analysis and recognition of rubbing fault.