Map Patterns in building groups embody the relationship of the material form of cities and their social-economic function, and reflect the city’s spatial structure, which have great importance in Cartographic Generalization and Multi-Scale Representation. On the basis of related research home and abroad, a Graph Theory approach is proposed to recognize the grid pattern in building groups. Firstly the cognitive characteristics and definitions of grid pattern are analyzed. Then neighborhood relationship is captured by proximity graph with the help of Delaunay triangulation, and visual distance model is established considering the Gestalt principles. Multi-connected linear pattern is recognized with constrains such as linear arrangement and compactness. Finally the line pattern intersection graph and similar orientation graph are formed. The candidate grid patterns are extracted through graph operation such as finding maximal complete sub-graph, join, intersection and post-processing. Experiments show that this approach is effective, feasible and practicable for obvious grid pattern recognition with the agreement of human spatial cognitive characteristics.