We conducted a meta-analysis to determine whether the within-person self-efficacy/performance relationship is positive, negative, or null and to compare the strength of the self-efficacy/performance and past performance/self-efficacy within-person relationships. The self-efficacy/performance within-person corrected correlation was .23 but was weak and nonsignificant (ρ = .06) when controlling for the linear trajectory, revealing that the main effect was spurious. The past performance/self-efficacy within-person corrected correlation was .40 and remained positive and significant (ρ = .30) when controlling for the linear trajectory. The moderator results revealed that at the within-person level of analysis: (a) self-efficacy had at best a moderate, positive effect on performance and a null effect under other moderating conditions (ρ ranged from –.02 to .33); (b) the main effect of past performance on self-efficacy was stronger than the effect of self-efficacy on performance, even in the moderating conditions that produced the strongest self-efficacy/performance relationship; (c) the effect of past performance on self-efficacy ranged from moderate to strong across moderating conditions and was statistically significant across performance tasks, contextual factors, and methodological moderators (ρ ranged from .18 to .52). Overall, this suggests that self-efficacy is primarily a product of past performance rather than the driving force affecting future performance.