Semantic feature modeling is an important research topic in CAD, in which geometric constraints in model are solved automatically. A semantic feature modeling method is given in this paper. Firstly, the feature dependent graph is built based on geometric constraints. Secondly, the feature dependent graph is decomposed according to the complexity of subgraphs and the whole problem of solving geometric constraints is divided into several small ones. Thirdly, these small problems are solved. At the same time, the modeling architecture based on the decomposition of feature dependent graph is given. Experimental results show that when the proposed method is applied, the modeling performance is improved.