Abstract: Because spatial structure has multi degree of freedom and complicated nonlinearity during the earthquake, the damage detection of spatial structure is difficult. A method of spatial structure damage detection under earthquakes is proposed by wavelet packet analysis in this paper. Furthermore, data fusion and clustering analysis are used. After constructing the damage index and diagnosis data bases, the damage location is detected using the minimum distance principle of clustering analysis. Noise effect is also studied, and the results show that noise has little impact on the accuracy of the damage detection method. The method is proven to be effective through numerical simulated analysis of a latticed shell.