The growing prominence of high-stakes exit examinations has made questions about their effects on student outcomes increasingly important. We take advantage of a natural experiment to evaluate the causal effects of failing a high-stakes test on high school completion for the cohort scheduled to graduate from Massachusetts high schools in 2006. With these exit examinations, states divide a continuous performance measure into dichotomous categories, so students with essentially identical performance may have different outcomes. We find that, for low-income urban students on the margin of passing, failing the 10th grade mathematics examination reduces the probability of on-time graduation by eight percentage points. The large majority (89%) of students who fail the 10th grade mathematics examination retake it. However, although we find that low-income urban students are just as likely to retake the test as apparently equally skilled suburban students, they are much less likely to pass this retest. Furthermore, failing the 8th grade mathematics examination reduces by three percentage points the probability that low-income urban students stay in school through 10th grade. We find no effects for suburban students or wealthier urban students.