We analyze the effect of election promises on electoral behavior in a laboratory experiment. In the experiment, politicians can make nonbinding election promises about how to split an endowment between themselves and the group. We find that promises affect both voting and voter beliefs about how much the politician will contribute to the public fund. The relationship is inverted U‐shaped with decreasing credibility of higher promises. Contributions of politicians are correlated with their promises in a similar pattern. The election promises are generally credible unless particularly high. Politicians keep promises more often if a reelection is possible and if the politician came into power by vote rather than by random draw. Voters reward high contributions in the previous period and punish promise breaking even after controlling for the contribution in the previous period or voters' beliefs about future contributions. By controlling for voters' beliefs, we distinguish retrospective from prospective voting. Our results suggest that voters both use promises for prospective voting and retrospectively punish broken promises.