In the current study, we empirically examined the proposition that failure experiences (a) alter the nature of the relationship between self-efficacy and performance and (b) shape the trajectory of self-efficacy over time. Participants (N = 128 young adult males) were randomly assigned to an induced failure or a control condition where they completed five sessions of a complex computer-based performance task, each preceded by a measure of self-efficacy. Multiple group discontinuous latent growth modeling (MGLGM) and piecewise hierarchical linear modeling (HLM) were used to compare similarities and differences across conditions in the self-efficacy–performance relationship and in the trajectory of self-efficacy over time. The findings showed that self-efficacy decreased immediately following induced failure; an effect that endured in later performance sessions, even after typical performance levels had recovered. Within-person (i.e., intra-individual) self-efficacy was negatively related to performance in the control condition, yet was unrelated to performance in the failure condition. Furthermore, the self-efficacy–performance relationship was dynamic such that between-person (i.e., inter-individual) self-efficacy was positively related to the trajectory of performance following failure, even when controlling for the effects of prior performance and task-related experience. These results suggest that failure experiences can fundamentally alter the form and direction of the self-efficacy–performance relationship.