Abstract: Spatial weights matrix is used to represent geographical feature similarity. In situ similarity is represented by different kinds of distance measurements based on Euclidean distance. This kind of similarity builds on spatial dependence but neglects spatial nonstationarity. In this paper, geographical feature similarity is defined as attribute and spatial similarity. We propose the concept of accumulated similarity surface and bring in curve evolution and fast marching method to calculate the accumulated similarity surfaces of geographical features. Spatial weights matrix is constructed using accumulated similarity surfaces according to both spatial dependence and spatial nonstationarity. Experiments are processed using trend surface and ASTER DEM as experimental data. The results show that spatial weights matrix based on accumulated similarity surfaces performs better than Euclidean-distance-based spatial weights matrix.