This study investigated the role of metacognition in event-based prospective memory. The aim of the study was to explore the relation between an item-level prediction (judgments of learning, JOL) and actual performance. The task and JOLs allowed a differentiation of the two components of prospective memory tasks (retrospective vs. prospective). Results revealed that individuals' predictions were (moderately) accurate for delayed JOLs but not for JOLs that had to be given immediately after task encoding. Moreover, data revealed an underconfidence-with-practice effect only for the retrospective component. For the prospective component, a substantial and general level of underconfidence in individuals' prediction–performance ratios was observed. The importance of metacognitive factors for prospective memory is discussed.