This study evaluated the transfer of training and training efficiency of two virtual reality environments (head-mounted display and personal computer) for a complex manual assembly task. Transfer of training was measured by comparing the post-training performance of two virtual training groups, a real-world training group and a control group that received no training. All training groups were taught to assemble a Lego(TM) forklift model in their respective environment. After training, participants assembled a real-world model of the forklift as well as a novel model of a racecar, which required the same parts as the forklift assembled in a different configuration. Results from this study show that virtual environments can be effective training simulators for complex assembly tasks although they are less efficient than real-world training. The results also suggest that individual differences such as general intelligence, spatial aptitude, and computer user self-efficacy influence one's ability to learn in a virtual environment.