In this paper we study electoral systems in an experimental governance game in which citizens contribute to a public good and policymakers decide how to distribute it. In the Voting condition, citizens receive information about the policymaker performance (how much she shared of the public good with individual participants), her skills and her altruism, and vote directly for a policymaker for the next election cycle. In the Referendum condition voters receive information about the policymaker's performance before deciding whether to remove her from office in a simpler binary choice (in or out). If they decide to remove the incumbent, a new policymaker is chosen randomly from the remaining group members, in the spirit of referendums like Brexit or the plebiscite about the peace agreements in Colombia. We compare these two electoral mechanisms with two baselines in which the policymaker is never replaced (Baseline) and another one (Random) in which the policymaker is always randomly replaced by another participant. Our results show that both electoral mechanisms are largely effective in improving citizens’ earnings by generating more equitable sharing rules. Citizens in Voting and Referendum extract greater surplus from policymakers than in the other conditions and voting against a candidate (Referendum) is more effective in increasing surplus than voting for a candidate (Voting). Contributions to the public good and conditional cooperation patterns remain strikingly similar in democratic (Referendum and Voting) and non-democratic (Baseline and Random) conditions. Empirical expectations about contributions of other citizens are similar as well. However, we show that policymakers distribute the public good differently in democratic institutions, and more effectively in Referendum than in Voting, as policymakers use clientelist transfers to avoid being replaced. Overall, both democratic institutions are effective in improving citizens’ payoffs by indirectly limiting policymakers’ surplus.