We evaluated two hypotheses-transfer appropriate monitoring (TAM) and the accessibility hypothesis-that explain why the accuracy of metacomprehension judgments is commonly low. In 2 experiments, participants read six expository texts, made global judgments about how well they would perform on a test over each text, and made term-specific judgments for predicting the recall of definitions embedded in each text. Criterion tests involved term-cued recall of the definitions. In Experiment 1, some participants made judgments after reading the texts, whereas others overtly attempted retrieval of each definition before making judgments. In Experiment 2, all participants had pre-judgment recall, and some also scored the correctness of the pre-judgment responses. Accuracy was greater for term-specific than global judgments, but only when pre-judgment recall was required. Term-specific accuracy was also constrained by accessing incorrect information. We argue that TAM is not a viable explanation of accuracy and discuss how to improve judgment accuracy.