We evaluate the labor-market returns to General Educational Development (GED) certification using state administrative data. We develop a fuzzy regression discontinuity (FRD) method to account for the fact that GED test takers can repeatedly retake the test until they pass it. Our technique can be applied to other situations where program participation is determined by a score on a "retake-able" test. Previous regression discontinuity estimates of the returns to GED certification have not accounted for retaking behavior, so these estimates may be biased. We find that the effect of GED certification on either employment or earnings is not statistically significant. GED certification increases postsecondary participation by up to four percentage points for men and up to eight percentage points for women.