For upper limb amputees, learning the control of myoelectric prostheses is difficult and challenging. Introduction of newer prostheses with multiple degrees of freedom controlled by various neural commands will make such training even more difficult. To produce smooth and human-like movements, the user must learn to produce multiple neural commands with precise amplitude and timing. To aid in training of the amputee users, we have developed a realistic and motivating virtual environment (VE) consisting of a physics-based target shooting game. The users' neural commands such as EMG, cortical neural activity, or voluntary movements of the residual limbs can be used to control the movement of a simulated prosthesis to point and shoot at virtual targets. In addition to the visual, sound, and performance feedback of the resulting movement, the game provides reaction forces in contact points that can be used to drive haptic displays. The timing measurements show that the physics-based simulation and rendering can be executed in real time in readily available PC systems. The target shooting game was developed in musculoskeletal modeling software (MSMS) that has been developed in our laboratory and is freely available for development of similar virtual training applications.