Illusions in judgments of learning (JOLs), predictions of future memory, occur when people base their JOLs on invalid cues or fail to consider cues that are predictive of actual memory performance. Several illusions have been reported (e.g., font size illusion, stability bias), but there is currently no established method to correct them. In this study, we tested the effectiveness of different forms of feedback in mending illusions. In each of three experiments, participants completed three study-test cycles in which they studied three different study lists. Participants made JOLs and received feedback or no feedback after each cycle. In Experiments 1 and 2, cognitive feedback on one’s own recall performance and JOL for each studied item was provided. In Experiment 3, additional metacognitive feedback informed learners about possible metacognitions during learning, their biased nature, and ways to enhance JOL accuracy. Results showed that cognitive feedback was not effective for mending the font size illusion (Experiments 1 and 2), the stability bias (Experiment 1), or the font format illusion (Experiment 2). In contrast, metacognitive feedback remedied the stability bias and in turn improved relative JOL accuracy (Experiment 3). In conclusion, this study shows that cognitive feedback alone at the item level is not enough for improving the cue basis and relative accuracy of JOLs. Rather, metacognitive feedback consisting of an in-depth explanation of biased metacognition is required.