The authors conducted a field experiment to explore why firms pay dividends. The authors change managers’ perception of agency concerns from outside investors, investors’ risk preference, the information gap with outside investors, and firms’ tax clientele by contacting publicly listed firms in China to test four dividend theories (agency, bird-in-hand, signaling, and tax clientele theories). The authors find that past payers (firms that paid dividends in the previous year) receiving the treatment of agency concerns increase their dividends relative to the control group. In contrast, firms receiving the other treatments do not experience changes in their dividend policy. The treatment effect of agency concerns in past payers is more prominent for firms with weaker governance and robust to various model specifications. A post-experimental survey confirms our treatment effects. The evidence suggests that the agency cost motive is the main determinant of a firm’s dividend policy.