Multi-scale representation of spatial data is a research focus in GIS, while building multi-scale data model is a key to implementing multi-scale representation of vector data. In view of the shortcomings of existing multi-scale data model in geographical cognition and special analysis, this paper puts forward a method of feature-based, and studies on it qualitatively from definition, description and extraction. Compared to conventional single-scale E-R model, in this paper, the key strategies of building multi-scale conceptual model are put forward. Deeply study and analysis are applied on abstraction and expression of multiple geometric characteristics, of multi-attribution, and of semantic relation among different scales,and the design of feature-based multi-scale conceptual model is realized. Finally, the object-oriented multi-scale logic model is researched, which lays a theoretical foundation for building the feature-based multi-scale vector data model.