As is well known, the extraction of the extraordinary signal, embedded in high-level white noise, is usually very difficult. In this paper, we suggest a new detection method for extracting such extraordinary signal by using wavelet transform, which we call Wavelet Pattern Matching. Our approach proceeds as 1) transferring signals from one dimension of the time series to two dimensions of the time-frequency plane by using wavelet transform, and analyzing the frequency characteristics of the signals ; then making a template based on the characteristics of the signal. 2) analyzing the objects and extracting the extraordinary signals by using Wavelet Pattern Matching. The method is proved to be effective when a model signal is used to test its effect. The effectiveness of this method becomes clearer when it is applied to extract the braking sound of an automobile embedded in high-level noise and knocking detection.