We conduct a public good experiment with high school teenagers. Some groups exclusively consist of students that we know to be friends. Other groups exclusively consist of students that we know not to be friends, and that are mere classmates. We find that ‘friends’ contribute more to the public good than ‘classmates’ do. Contributions of ‘classmates’ sharply decrease in the last round, in line with the literature on public good experiments. However, contributions of ‘friends’ sharply increase in the last round.