This study investigated the heuristic bases of judgments of learning (JOLs). JOLs were elicited either immediately after study or after a shorter or longer delay. In Experiment 1, the effects of encoding fluency (inferred from self-paced study time) on both JOLs and recall decreased with JOL delay, whereas those of retrieval fluency (inferred from the success and latency of pre-JOL retrieval) increased. In this experiment, JOLs (as well as recall) decreased with increasing study time, presumably under the heuristic that items requiring more time to study are less likely to be recalled. In contrast, in Experiment 2, in which study time was experimentally manipulated, JOLs as well as recall actually increased with study time. In both experiments JOLs increased with retrieval fluency. The results demonstrate that JOLs are based on the flexible and adaptive utilization of different mnemonic cues according to their relative validity in predicting memory performance.