This study investigated whether individuals can predict their future prospective memory (PM) performance in a lab-based task and in a naturalistic task. Metacognitive awareness was assessed by asking participants to give judgments-of-learning (JOLs) on an item-level for the prospective (that something has to be done) and retrospective (what to do) PM component. In addition, to explore whether giving predictions influences PM performance, we compared a control group (without predictions) to a prediction group. Results revealed that giving predictions did not change PM performance. Moreover, participants were underconfident in their PM performance in the lab-based task, while they were overconfident in the naturalistic task. In addition, item-level JOLs indicated that they were inaccurate in predicting what items they will recall or not, but only for the prospective component of the PM task. As for the retrospective component, they were equally accurate in both task settings. This study suggests a dissociation of metacognitive awareness for PM according to both task setting and processing component.