In the process of dynamically monitoring bridges using GPS, there are unavoidably errors (noises) in the observation data, for example the effect of multi-path effects. These noises possibly cover up available signals that we need. The paper analyzes field survey data of GPS dynamic monitoring for a bridge. Two GPS double-frequency receivers (Trimble 5700) were used for collecting data about 20 minutes by means of the dynamic observing mode. The baseline for monitoring the bridge deformation is resolved based on the least-square adjustment, in order to get the coordinate series of the monitoring baseline, and then the data series of three dimension coordinates, which are the rover station relative to the reference station, were calculated. The standard deviation is about 10mm in horizontal and 20mm in elevation. After averaged, these series were decomposed to eliminate the high frequency noises using ddencmp function of wavelet packet in the MATLAB6.5 Toolbox, in which a threshold was made. And then the signals were rebuilt through wdencmp function. The results have shown that noises can effectively be removed and the useful signals can be extracted from original signals with wavelet analysis. Further, the bridge self-vibration frequency can be obtained by using the spectral analytic approach with Fourier transformation, which is very identical with theory calculating value that was made by other researchers before, and more precise than the results without using wavelet