Virtual reality has been used widely as a computerized medium for training. In addition to visual and auditory feedback, the addition of haptic feedback to virtual environments envisions promising applications for human motor skill acquisition, such as rehabilitation and training of surgeons. Generally speaking, there exist two kinds of virtual training paradigms, virtual practice and haptic guidance. Compared to virtual practice, haptic guidance conveys more information to the trainee, some of which might not be realizable in the physical world, in order to improve the training efficacy over practice. In this thesis, the most general form of haptic guidance, shared control, is introduced. A shared controller dynamically intervenes, through an automatic feedback controller acting upon the system, to modify the coupled system dynamics during training. The coupled system dynamics are to be selected to help expedite learning of the task. Specifically, an error-reducing shared controller is proposed and implemented for this dissertation. A series of human-participant studies have been conducted to test the efficacy of the proposed shared control approach for performance enhancement and training in virtual environments. During the experiments, performance is measured utilizing a Fitts'-like under-actuated target-hitting task. Experimental results indicate that an error-reducing shared control paradigm with fixed control gains does enhance performance of the participants during the target-hitting task when assistance is on. However, a one-month long training experiment reveals that the error-reducing shared control paradigm with fixed control gains exhibits negative efficacy during training in virtual environments. It is suggested that the fixed control gains implemented in the shared controller are one of the primary reasons for negative efficacy of haptic guidance for training. Therefore, a shared controller with performance-based gain adaptation was also tested. In addition to comparisons with the fixed-gain shared controller, this paradigm is also compared to virtual fixtures, another commonly used form of haptic guidance. The experimental results indicate that the performance-based progressive shared control paradigm results in significantly better training performance than all other haptic guidance training protocols implemented in this study. Hence, the training effectiveness of performance-based error reducing shared control is verified. Finally, this study is aimed to improve the design of progressive shared control algorithms for manual control tasks. Therefore, a perception study has been conducted quantifying human sensitivity to varying system dynamics. The results of this study form the basis for properly choosing the step size of the progressive control gains of shared controllers.