Vibration of the bridge always carries information about its function. But in general, the vibration contains many noises. In order to obtain correct information, the background noise must be removed or the vibration must be purified. A de-noising method is given in this paper and is successfully used in feature vibration extraction. We present a method on wavelet transform to eliminate the noise. It is difficult to detect an instantaneous frequency spectrum throughout the sampling period by the FFT. Since the FFT provides an averaged frequency by spectrum of the total sampling period, the CWT can be used for discovering the signal components, using a proper basic wavelet. We can obtain the feature components of a signal by reconstructing the wavelet coefficients. This method is used for extracting the vibration of the bridge with different types of failure. The feature vibration is extracted successfully.