A possible paradox arises from two major paradigms in the literature studying transfer: designing training to increase transfer (e.g., by including task variation) may lead to lower self-efficacy. The present study investigated this paradox by examining the relationships among design of the practice condition, self-efficacy, and transfer. 82 participants (36 men, 46 women) filled out premeasures, trained on a computer-based task, filled out a self-efficacy measure, and completed a computer-based task for the transfer test. The practice condition was found to impact transfer performance, though not in the expected direction for all transfer tests. While the practice condition did not impact self-efficacy level, the practice condition did impact self-efficacy generalization. Weak support was found for a relationship between self-efficacy and transfer performance. These results suggest relationships among practice condition, self-efficacy, and transfer. Specifically, they indicate that the design of training can influence not only transfer, but also self-efficacy generalization.