Student learning in introductory science, technology, engineering, and mathematics (STEM) courses is often self-regulated. For self-regulated learning to be effective, students need to engage in accurate metacognitive monitoring to make appropriate metacognitive control decisions. However, the accuracy with which individuals monitor their task performance appears to largely overlap with their ability to perform that task. This study examined the trajectories in the accuracy of students' metacognitive monitoring over the course of a semester, along with the effect of monitoring accuracy feedback. The results indicate that some students improve the accuracy of their predictions over the course of a semester. However, low-performing students are less accurate at predicting their exam grades, and tend not to improve their metacognitive calibration over the course of a semester. In addition, providing low-performing students with calibration feedback may lead to greater overconfidence.