Interaction is one of the important features of virtual assembly. In order to promote the system s interactive performance, computer and human should share in cognitive burden in reason. The realization of virtual environment s perceptual mechanism is the critical method towards this objective. In this paper, a formal description of virtual assembly structure is given first, then a perceptual mechanism of compound object is put forward, lastly, a three-stage perception mode is given, and the perceptual process constructions which are needed in virtual assembly application are also analyzed. The desktop virtual assembly practices demonstrate that this perception mechanism has a good real-time interactive characteristic, and it is also a wonderful facility to help understanding user's intention.