Through analysis of Traditional CAD software modeling methods, combined with process planning requirements of intelligent virtual assembly, in the proposed semantic-based multi-level top-down model of virtual assembly. This paper proposes a semantic-based top-down model of multi-level virtual assembly.On this basis, the assembly semantic information extraction process is Focused analyzed, in the three-dimensional model data, non-geometric factors such as assorted features, assembly features, assembly engineering relations are extracted, and modeling is Established by Protégé ontology-based tool, the model is imported into the 3DVAPP(3D Virtual Assembly Process Planning) system to be verified, and Finally this paper implementated the assembly process planning applications such as intelligent parts identification, Provide a theoretical and practical basis for intelligent virtual assembly.