This paper argues that assortative matching may explain over-education. Education determines individuals' income and, due to the presence of assortative matching, the quality of the partner, who can be a colleague or a spouse. Thus an individual acquires some education to improve the expected partner's quality. But since everybody does that, the partner's quality does not increase and over-education emerges. Tax progression to correct over-education has ambiguous effects on the educational incentives according to the individuals' ability. We test the model using the British Household Panel Survey. The empirical results support our theoretical findings.