This research explored whether academic self-efficacy calibration (the match between self-efficacy beliefs and academic outcomes) in first-year psychology students (n=197) differed as a function of task type (written assignment/multiple-choice exam), domain specificity (task level/subject level), over time (mid-semester/end of semester) and according to student achievement level (high achievers/low achievers). Lower-achieving students were overconfident across both the written assignment and the exam, while higher-achieving students were accurately calibrated on both tasks. The subject-level calibration of lower-achieving students improved between mid-semester and the end of semester (though students remained overconfident). Higher-achieving students' subject-level calibration remained stable over the semester, and they were about half as overconfident as the lower-achieving students. Both groups of students were more overconfident at subject-level than at task-level overall. On the whole, overconfidence was prevalent, especially for low achievers, and at subject level. Findings suggest that a one-size-fits-all approach to self-efficacy is unlikely to be beneficial for all learners.