Serious games are becoming increasingly popular due to their association with increased learning outcomes when compared to traditional self-regulated learning activities. However, the majority of research examining the outcomes of serious games has focused almost exclusively on learning outcomes. This has resulted in a lack of research examining why these types of games result in increased positive outcomes, such as engagement or performance. This study seeks to address this gap in existing research by examining the relationship between game difficulty and participants' engagement, performance and self-efficacy in a Pacman style maze navigation game. This required the use of hidden difficulty variations which participants were randomly assigned. Participants engaged with the game over a 5-days practice period. Results from this study suggest that difficulty plays a considerable role in influencing participants' self-efficacy for the task. Self-efficacy has been consistently linked to positive outcomes such as increased engagement and performance. This highlights the importance of difficulty as a game design factor as well as providing an insight into the manner in which serious games could be further refined in order to increase user's self-efficacy and associated positive outcomes. Implications for future serious games and self-efficacy research are discussed.