The invention discloses an improved PCA damage detection method under a spatiotemporal window. Firstly, a numerical simulation or experiment means is used to determine a spatial influence range corresponding to damages at different positions under multiple working conditions with damages at different spatial positions, a structural space area sensitive to the damages is further determined, and that is, a spatial window is determined; then, according to a PCA principle, a temporal window related to an external load cycle is determined; and finally, after the temporal window and the spatial window are determined, a PCA method is used for all measurement point data in the spatial window, a feature vector is used as a monitoring index, a damage can be monitored in real time, and the damage positioning is realized. The method carries out PCA analysis on a structural response in limited temporal and spatial windows, the damage recognition precision and the sensitivity are thus improved, andlong-term structure monitoring data can be used to carry out damage recognition on a long-span solid bridge or other structures.