In this paper, the basic theory and core techniques of AVR (from Actual Reality to Virtual Reality) have been expounded. AVR is a development of VR technique. In AVR, the virtual objects in virtual environment are modeled directly from the shape information of their counterparts in real world based on computer graphics and computer vision techniques, which provides the virtual environment with a higher level realism and preciseness. In order to achieve the integration of computer vision and computer graphics, superquadric based (SQ-based) model as the universal description of virtual objects is chosen and PAMVISION (Part-based Multiple-layer Visual Information Organization Model) model providing a general framework for vision recognition and reconstruction in AVR modeling is developed. As the core techniques of AVR, two approaches for virtualizing the real objects from either 2D images or 3D data are im plemented. As an application of AVR, an interactive 3D superquadric based modeling software package is developed for creating virtual environment.