The use of computer-based learning (CBL) systems by information systems educators is rapidly growing. While improvements in student computer skills test results have been attributed to the use of such systems, little is known about the theoretical mechanisms that may be contributing to such improvements, or whether all students benefit equally from all CBL system training features. In this study, we explore self-efficacy theory as a framework for understanding how CBL systems influence student computer performance. More specifically, we examine the effectiveness of CBL systems in raising efficacy beliefs via two sources of efficacy information - enactive mastery and vicarious experience. Preliminary results revealed that students with lower initial specific computer self-efficacy (SCSE) beliefs benefited more from vicarious learning features that demonstrated concepts, whereas those with higher initial SCSE beliefs benefited more from enactive mastery features in which they could experiment on their own. Moreover, post training SCSE judgments were significantly related to computer skills testing scores. Based on our findings, educators can more precisely match CBL features with student demographics such as initial SCSE perceptions, and in turn, improve downstream student computer skills testing performance.