Abstract: In the absence of semantic information, the selection of point group is one of the difficulties in cartographic generalization. This paper proposes a new multi-level clustering point group generalization method which takes into account density contrast. Firstly, in view of the shortcoming of k-means clustering algorithm, this paper uses an improved density peak clustering method to realize automatic clustering of point group, mainly reflects on determining the optimal cut-off distance by the Gini coefficient and uses the relation of local density and relative distance to detect the clustering centers. Secondly, we propose a point group selection strategy which takes into account density contrast, the point group is divided into clusters of different grade by multi-level clustering. The clustering centers of different grades are determined, and the hierarchical tree structure of point group data is established. The number of points to be selected is calculated according to the square root law, then allocated from top to bottom according to the number of clusters at each level, and the selected objects are determined, and automatic selection of points and multi-scale expression of point group are realized. Point groups experimental results with different distribution patterns show that the method described in this paper can get reasonable selection results, which verifies the universality and effectiveness of the method.