Abstract: During health monitoring of truss structures, structural integrity monitoring is limited by the number of sensors. To solve this problem, a POD-ME method based on proper orthogonal decomposition (POD) and modal expansion (ME) is developed. In this method, it is not necessary to obtain prior information by establishing a finite element model, and the proper modes reflecting the structure feature are trained using the measured dynamic response data only. The proper modes are chosen to construct a virtual sensing matrix that characterizes the mapping relationship between the sparse strain of truss and the strain at the unmeasured points, combined with the virtual sensing matrix and the dynamic response data of finite measuring points, and then an effective estimation of the truss structure from the strain response of finite measuring points to the strain response of unmeasured points is realized. The numerical simulation of a truss structure and a vehicle-passing truss bridge test are carried out to verify the effectiveness and robustness of the developed method. The influences of the number of POD modes, noise levels, different patterns of sensor placement and different excitation conditions on the estimated results are discussed. The results show that the unmeasured response of the structure is effectively estimated by the developed method under the condition that only some elements can be measured.