Judgments of learning (JOLs), as one type of metacognitive judgments, are assessments that people make about how well they have learned material. The effective use of JOLs depends on various factors, including task-specific variables and the learner's own metacognitive resources. Little has been known about the relationship between JOL accuracy for memory predictions and metacognitive ability, which is an emerging theory-practice gap in the field of metacognition. The present study investigated the relationship between the absolute accuracy of JOLs and the metacognitive awareness inventory (MAI), using concrete and abstract word pairs through three study-test cycles. We found that participants who scored high on the MAI, also produced a high level of absolute accuracy on each study-test cycle. The results from mediation analyses yielded that the impact of the MAI on absolute accuracy on cycle 3 was completely mediated by absolute accuracy on the first two cycles for both concrete and abstract word pairs. In addition, the same pattern of results was obtained even when a subset of the MAI (either knowledge or regulation of cognition) was used. These indicate that trait-based metacognitive abilities can well explain the correspondence between JOLs and recall performance on the first test. However, their impact is reduced after study-test practice, suggesting that experience-based factors become critical to improving metacognitive accuracy.