Students often gauge their performance before and after an exam, usually in the form of rough grade estimates or general feelings. Are these estimates accurate? Should they form the basis for decisions about study time, test-taking strategies, revisions, subject mastery, or even general competence? In two studies, undergraduates took a real multiple-choice exam, described their general beliefs and feelings, tracked their performance for each question, and noted any revisions or possible revisions. Beliefs formed after the exams were poor predictors of performance. In contrast, real-time metacognitive monitoring--measured by confidence ratings for each individual question--accurately predicted performance and were a much better decisional guide. Measuring metacognitive monitoring also allowed us to examine the process of revising an answer. Should a test-taker rely on their first choice or revise in the face of uncertainty? Experience seems to show that first instincts are correct. The decision-making literature calls this the first-instinct fallacy, based on extensive analysis of revisions, and recommends revising more. However, whereas revisions have been analyzed in great detail, previous studies did not analyze the efficacy of sticking with an original choice. We found that both revising and sticking resulted in significantly more correct than incorrect outcomes, with real-time metacognition predicting when each was most appropriate.