In the context of repeated public good contribution games, we experimentally compare the institution of democratic punishment, where members of a group decide by majority voting whether to inflict punishment on another member, with individual peer-to-peer and dictatorial punishment institutions. Democratic punishment leads to more cooperation and higher average payoffs, both under perfect and imperfect monitoring of contributions. A comparison with dictatorial punishment suggests that the effect relative to traditional peer-to-peer punishment primarily works by curbing anti-social punishment and thereby establishing a closer connection between a member's contribution decision and whether subsequently being punished by others.