When a large machine is working，the signal-to-noise ratio（SNR） of its vibration signal is very low and the frequency of characteristic signal is very high. The signal is overwhelmed by strong background noise, so it is difficult to pick it up. A new diagnosis method by using correlation analysis and wavelet transform was introduced. The correlation function has been used to suppress the noise and the wavelet for multi-resolution analysis. The effectiveness of suppressing noise and robust ability of detecting fault feature by using this method have been proved by both simulation and experiments done on a gun-automaton gearbox.