While it has been consistently demonstrated that academic self-efficacy and performance are positively correlated in groups of students, little is known about whether individual students' academic self-efficacy levels align with their own performance abilities. At the same time, researchers contest whether self-efficacy should align with performance abilities to be of most benefit to students. In this study, we applied procedures used in the meta-cognitive calibration paradigm to investigate the alignment between academic self-efficacy and academic performance (i.e., self-efficacy calibration) in higher education. Undergraduate students (n = 207) completed five self-efficacy questionnaires with regard to academic performance outcomes in one subject over a semester (two written assignments, two exams, and the subject overall). Five corresponding grades were also collected. We calculated two types of self-efficacy calibration scores: self-efficacy accuracy (the deviation between self-efficacy and performance) and self-efficacy bias (the signed difference [i.e., valence]; over- and under-efficaciousness). Miscalibration of self-efficacy beliefs was prevalent, consistent with findings regarding meta-cognitive calibration. Under-efficaciousness was common at task level (for written assignments and exams), while over-efficaciousness was pronounced at domain level (for the subject overall). Self-efficacy exceeded performance for low-achievers, while it fell short of performance for high-achievers. A key finding was that self-efficacy bias predicted academic performance on similar subsequent tasks, with under-efficacious students performing better than accurate or over-efficacious students. Findings suggest self-efficacy is not a self-fulfilling prophecy; instead, over-efficacious students may experience negative impacts on academic self-regulation and performance.