By recruiting a modular organization of muscle with relative activities, the arm motion can be indicated by the neural system and generated for performing a variety of motor tasks. In this study, a Non-negative Matrix Factorization with initial estimation is applied to identify and extract primary muscle synergies and their activation patterns from the processed EMG recordings during three multidirectional tracking tasks with visual feedback interaction. The effects of task variety and tracking accuracy by visual feedback on muscle synergies and their activation patterns are explored by statistic analysis. The results showed that only the task variety affected what synergies were indicated by the neural system, but both task variety and visual feedback affected the duration and magnitude of the primary synergies. Thus, for active rehabilitation application, it is advised that if the purpose is to enhance the synergy indication from the neural system, the task completion accuracy should be emphasized, but if the purpose is to expand the motion area, the task variety should be diversified.