An effective procedure is proposed for scanning and separating the vertical (flexural) and torsional–flexural frequencies of thin-walled girder bridges by a moving single-axle test vehicle, modeled as a two degree-of-freedom (DOF) system to account for the vertical and rocking motions. The response of the vehicle is not directly used, since it may mask the bridge frequencies by self frequencies. Instead, the wheel-bridge contact responses are used for being free of vehicle’s frequencies. To start, closed-form solutions are derived for the vertical, lateral, and torsional vibrations of the mono-symmetric beam. Then, the contact responses are calculated from the vehicle responses, considering the discrete nature of field data. The vertical and torsional–flexural frequencies of the bridge are separated by using the vertical and rocking motions derived from the contact responses without prior knowledge of the mode shapes. The proposed technique is numerically validated with the following conclusions: (1) the wheels’ contact responses outperform the vehicle response in that more high frequencies of the bridge can be detected; (2) the torsional–flexural frequencies can better be detected from the wheel closer to the bridge edge; and (3) the vertical and torsional–flexural frequencies of bridges can be successfully separated by the proposed procedure even in the presence of pavement roughness, which is helpful for field applications.