In order to properly address the resonance phenomenon which is one of the most typical vibration problems of railway bridges, it is important to identify the vibration characteristics of the railway bridges at the time of train passing, and also to quantify its uncertainty. In this study, the methodology which can identify the vibration characteristics and its uncertainty by employing the SSVS (Stochastic Search Variable Selection)-VAR (Vector AutoRegressive) model and the Bayesian estimation is proposed. By applying the actual acceleration response of the bride at the time of train passing, it was found that 1) natural frequency and modal damping ratio under train passing and after train passing could be identified stably, respectively, 2) the uncertainties of natural frequency and modal damping ratio can be identified as 2% and 50%, respectively.