Aimed to extract the relative energy of engine cylinder head vibration signals, a new method based on dual tree wavelet package transform and adaptive SURE block threshold de-noising was proposed. The DT-CWPT make sure that the coefficients decomposed by tree a and tree b can be complementary information at each scales, so the DT-CWPT can achieve approximate shift invariance and decrease the information missing. The adaptive block threshold de-noising method gains the best threshold and the optimal neighborhood effect length for signal de-noising and can obtain higher signal-to-noise. Also causes of frequency aliasing and frequency bands derangement were discussed and Fourier transform method was applied to suppress the false frequency in dual tree wavelet package transform. The simulated signal and measured signal test results show that the proposed method has a good de-nosing performance and more effective in fault feature extraction.