BACKGROUND Evidence of a negative relationship between educational attainment and divorce in Japan is not consistent with predictions derived from standard theoretical emphases on the costs of divorce. OBJECTIVE Using marital history data from a cross-sectional survey, we estimated educational differences in divorce for two marriage cohorts: 1980-89 and 1990-2005. We also used 14 years of panel survey data to evaluate four possible explanations for the observed negative educational gradient. RESULTS Our results confirmed that educational attainment is inversely related to divorce in Japan, and showed that, in contrast to some previous findings, the negative relationship between women's education and divorce has not become stronger in recent years. Analyses of the panel data provided some support for hypotheses that focused on the role of economic strain and on cultural values regarding reputation or "face," but they also showed that the negative relationship between education and divorce remained strong even after controlling for a range of posited correlates. CONCLUSIONS Our failure to solve the theoretical puzzle motivating these analyses suggests that other types of contextual modification to standard theories of family change are required to explain the strong negative relationship between educational attainment and divorce in Japan. We discussed possible examples of such modifications, focusing on the patterns of selection into marriage and the central importance of investment in children's educational success in Japan's highly competitive educational system, while also offering more nuanced theorization regarding the role of reputation or "face".