This paper presents a theoretical algorithm for constructing the mode shapes of a bridge from the dynamic responses of a test vehicle moving over the bridge. In comparison with those approaches that utilize a limited number of sensors deployed on the bridge, the present approach can offer much more spatial information, as well as higher resolution in mode shapes, since the test vehicle can receive the vibration characteristics of each point during its passage on the bridge. Basically only one or few sensors are required to be installed on the test vehicle. Factors that affect the accuracy of the present approach for constructing the bridge mode shapes are studied, including the vehicle speed, random traffic, and road surface roughness. Through numerical simulations, the present approach is verified to be feasible under the condition of constant and low vehicle speeds.