In judgments of learning (JOLs), participants predict whether they will retrieve an item in a memory test. JOLs rely on multiple cues, some of which are predictive of memory performance (e.g., number of presentations), some are not (e.g., font size). In judgment and decision making research, multiple cue probability learning studies have shown that people can learn to weigh judgment-relevant cues appropriately when they make judgments across multiple blocks with adequate feedback. In each of two experiments, participants completed three study-test cycles with JOLs. In Experiment 1 (n = 160), a typically overweighted invalid cue (font size) and a typically underweighted valid cue (number of future study presentations) were used. In Experiment 2 (n = 80), two invalid cues (font size, fOnT fOrMaT) were varied. In addition, the informativeness of feedback varied between no explicit feedback and very detailed "cognitive feedback" explaining the two cues to attend to, and the average number of words in each category remembered by former participants (or by themselves in Exp. 2) along with hints which features are typically over- or underestimated in JOLs. Whereas performance in the recall tests improved somewhat across study-test cycles, the accuracy of JOLs in terms of resolution was unaffected. An adequate reduction of the effect of font size on JOLs in Experiment 1 was not replicated in Experiment 2. Cue weighting was unaffected by study-test experience, even with highly informative feedback. Hence, we conclude that JOLs are resistant to improvement by experiencing feedback, even if it is optimally informative.