Building-structure type is a crucial indicator for assessing seismic performance. Traditional remote-sensing assessments of building seismic performance mainly exploit the visual features of individual buildings, while neglecting their spatial similarity. To enhance the classification accuracy of house structure types, this paper introduces a Minimum Spanning Tree (MST) clustering method based on the spatial similarities of houses. First, the geometric attributes of each building are used to compute a Gestalt factor that quantifies visual distance. A Delaunay triangulation is then built to establish a neighborhood graph, and an MST is generated with the visual distance as the edge weight. By pruning the MST, clusters of buildings with high spatial proximity are obtained. Finally, a support-vector machine (SVM) classifies each cluster into one of four structural types—simple, brick-frame, and frame-shear-wall—using geometric, textural, height and spatial features. The classification of housing clusters is divided into three categories: simple houses, brick-concrete houses, and frame houses. This article conducted remote sensing classification experiments on building structure types in Weining County, Guizhou Province. Compared with the seismic resistance classification based on single family houses, the classification accuracy improved by more than 10%; The kappa coefficient has increased from 0.58 to 0.79. The experimental results indicate that the feasibility of introducing the MST clustering method.