The objective of this study is to analyze the different role that efficacy beliefs play in the prediction of learning, innovative and risky performances. We hypothesize that high levels of efficacy beliefs in learning and innovative performances have positive consequences (i.e., better academic and innovative performance, respectively), whereas in risky performances they have negative consequences (i.e., less safety performance). To achieve this objective, three studies were conducted, 1) a two-wave longitudinal field study among 527 undergraduate students (learning setting), 2) a three-wave longitudinal lab study among 165 participants performing innovative group tasks (innovative setting), and 3) a field study among 228 construction workers (risky setting). As expected, high levels of efficacy beliefs have positive or negative consequences on performance depending on the specific settings. Unexpectedly, however, we found no time × self-efficacy interaction effect over time in learning and innovative settings. Theoretical and practical implications within the social cognitive theory of A. Bandura framework are discussed.