Judgments of learning (JOLs) made after a 5-min delay are almost perfectly accurate: the "delayed-JOL effect" (T. O. Nelson & J. Dunlosky, 1991). The mechanisms underlying this phenomenon have been the subject of debate. This study examined the effects of delays and short-term memory (STM) distraction on memory and metamemory (JOLs). STM distraction (2.5-30 s) immediately following encoding increased both JOL accuracy and mean cued recall. However, JOLs made after longer delays (4-5 min) were even more accurate. In addition, making a JOL at longer delays improved cued-recall performance. Conditional probabilities of cued recall (given successful initial retrieval) also increased over time and with interference, indicating that delayed JOLs may alter what they assess. Finally, increased confidence was associated with shorter JOL latencies only at delays. The results are consistent with an accessibility view of metamemory (e.g., A. Koriat, 1993).