This paper describes a novel approach to generate and model assembly semantic meaning enclosed in product assembly features in order to optimize the preparation time of Virtual Reality simulations. The proposed approach is based on a set of heuristic rules to generate semantic KeyPoints (characterisation of a kinematic link or a mates) used to idealize an assembly model. This study iden- tifies through a disassembly process a number of semantic rules in order to extract and translate assembly semantic features from CAD models. The proposed ap- proach is based on two steps: features extraction and semantic recognition of the assembly features. In the first step, internal boundary representation (B-Rep) and mate extraction methods are used to retrieve the engineering meaning from as- sembly models using SolidWorks’ API functions. In the second step, a multi- level semantic rules model is used. The approach is demonstrated and validated on a use-case with a disassembly process scenario and adapted to Virtual Reality.