Envelope analysis is popular and effective for bearing fault diagnosis. Traditionally, the frequency band enclosing a structural resonance is selected manually for envelope analysis. In this paper, a method determining this band automatically is developed. Wavelet packet transform is applied to decompose bearing signal into different frequency bands with equal bandwidth. RMS values of these bands are calculated and the band with the biggest RMS is picked up for envelope analysis. Further, cepstrum is used to process modulating signal to acquire bearing fault signature automatically. Experimental results confirm the satisfactory performance of these two techniques.