For detecting the bearing fault of the brushless DC motor at the early state, this paper proposes a new method to abstract the bearing fault information. The proposed method chooses the bus current and the forward envelope of three phase current as the signal extraction object， avoiding the disadvantages of the phase current for its rich harmonic and discontinuous. Meanwhile, current signal is decomposed by wavelet package transformation. In this way, the bearing fault information can be abstracted clearly. The experiment result shows that the bearing fault information can be abstracted from the bus current and the forward envelope of three phase current by wavelet package transformation.