We construct a model of household decision-making in which agents consume a private and a public good, interpreted as children's welfare. Children's utility depends on their human capital, which is produced from parental time and human capital. We first show that as returns to human capital increase, couples at the top of the income distribution should spend more time on children. This in turn should reinforce assortative matching, in a sense we precisely define. We then embed the model into a Transferable Utility matching framework with random preferences a la Choo and Siow (2006) which we estimate on US marriage data for individuals born between 1943 and 1972. We find that the preference for assortative matching by education has significantly increased for the white population, particularly for highly educated individuals; but not for blacks. Moreover, in line with theoretical predictions, we find that the "marital college-plus premium" has increased for women but not for men.