Guided by a hypothesis that integrates principles of monitoring from a cue-based framework of metacognitivejudgments with assumptions about levels of text representation derived from theories of comprehension, we discovered that rereading improves metacomprehension accuracy. In Experiments 1 and 2, the participants read texts either once or twice, rated their comprehension for each text, and then were tested on the material. In both experiments, correlations between comprehension ratings and test scores were reliably greater for participants who reread texts than for participants who read texts only once. Furthermore, in contrast to the low levels of accuracy typically reported in the literature, rereading produced relatively high levels of accuracy, with the median gamma between ratings and test performance being +.60across participants from both experiments. Our discussion focuses on two alternative hypotheses-that improved accuracy is an artifact of when judgments are collected or that it results from increased reliability of test performance-and on evidence that is inconsistent with these explanations for the rereading effect.Accurately assessing one's text comprehension is instrumental in effectively learning new material. Accordingly, many researchers have investigated the theoretical bases of comprehension assessments, or metacomprehension, with one goal being to discover conditions that produce high levels of metacomprehension accuracy. In most metacomprehension research, participants read several short texts, rate their comprehension for each text, and then complete tests for each. With few exceptions, metacomprehension accuracy has been poor. For instance, Maki (1998b) reported that the mean correlation between ratings and test performance across 25 studies from her lab was only .27. On the basis ofa review ofthe entire literature, Maki (1998b) concluded that "the low accuracy of text predictions may mean that students cannot predict performance well. Alternatively, low accuracy may indicate that our measurement of metacomprehension accuracy is too unreliable for us to detect changes" (p. 142).Given such poor accuracy found in previous research, a question naturally arises: Can metacomprehension accuracy be improved? We provide an affirmative answer to this question by integrating current theory ofmetacognitive monitoring with theory of text comprehension.