Objective This article documents trends in the educational gradient in marriage in Japan and evaluates the mechanisms underlying those trends by drawing on two broad theoretical frameworks emphasizing gender revolution and labor market bifurcation. Background Efforts to explain reversal of the negative educational gradient in women's marriage have been limited in both theoretical scope and geographical focus. Distinctive features of the Japanese context allow us to shed new light on multiple sources of change in marriage behavior. Method The authors apply the harmonic-mean two-sex marriage model to person-period data for men and women to investigate the changes in marriage rates by educational attainment and by specific educational pairings during the period 1988 to 2015. Our focus on changes in patterns of educational pairing facilitates the evaluation of predictions from the two broad theoretical frameworks. Results The analyses showed that the negative educational gradient in women's first marriage disappeared by 2005 and that a positive gradient emerged after 2009. This change was brought about by a combination of decline in the marriage rates of less-educated women and an increase in the marriage rates of highly educated women. The former was the result of declining educational homogamy, whereas the latter was due to a rise in female educational hypogamy. In contrast, the educational gradient for men remained positive throughout the time period examined. Conclusion Our study provides the first evidence of a positive educational gradient in women's marriage in Japan and provides empirical evidence consistent with expectations derived from two different, but interrelated and potentially reinforcing, theoretical frameworks.