<p>Structural damage detection is hard to conduct in large-scale civil structures due to enormous structural data and insufficient damage features. To improve this situation, a damage detection method based on model reduction and response reconstruction is presented. Based on the framework of two-step model updating including substructure-level localization and element-level detection, the response reconstruction strategy with an improved sensitivity algorithm is presented to conveniently complement modal information and promote the reliability of model updating. In the iteration process, the reconstructed response is involved in the sensitivity algorithm as a reconstruction-related item. Besides, model reduction is applied to reduce computational degrees of freedom (DOFs) in each detection step. A numerical truss bridge is modelled to vindicate the effectiveness and efficiency of the method. The results showed that the presented method reduces the requirement for installed sensors while improving efficiency and ensuring accuracy of damage detection compared to traditional methods.</p>