Left: Sleep subjects with verbal reports related to the maze improved tenfold more at retest than did participants without task-related mentation. In contrast, thoughts about the task during Wake did not provide a similar benefit. Error bars = SEM. Right: Baseline performance was a strong predictor of later improvement (regression lines and 95% CI lines for all subjects). Sleep participants reporting maze-related dreams (n=4, large circles) were amongst those with the poorest baseline performance, but improved significantly more at retest than other poor performers. In contrast, subjects who reported thoughts of the maze task during Wake (n=2, large circles) did not differ from others in terms of baseline performance, and improved similarly to those with comparable baseline performance. See also Supplemental Figure S1.