Abstract -- From measurement signals, the characteristic of structure system can be extracted by using signal processing techniques. The most important characteristics of structures are including modal natural frequency and modal damping ratio. Such an inverse operation to figure out the parameters is called system identification. In reality, the modal parameters are not necessarily constant with time. The traditional system identification techniques based on constant parameters or stationary assumptions are not adapted to non-stationary cases. And new identification techniques for general non-stationary system are developed. The project focuses on identifying bridge structures by using wavelet analysis and further applying to structure health monitoring. Wavelet transform can identify an instantaneous spectrum and give a multi-resolution analysis. Choosing a proper mother wavelet, the feature components of signal can be obtained by reconstructing the wavelet coefficients. Based on feature components, we develop a series of criteria to judge the degree and locations of damages.