De-noising and extraction of the weak signal are very important to mechanical fault detection in which case signals often have very low signal-to-noise ratio (SNR). In this paper, a denoising method based on the optimal Morlet wavelet is applied to feature extraction for mechanical vibration signals. The wavelet shape parameters are optimized based on kurtosis maximization criteria. The effectiveness of the proposed technique on the extraction of impulsive features of mechanical fault signals has been proved by practical experiments.