How far does democracy decrease corruption? And which specific aspects of democracy help generate such effects? Corruption is famously one of the strongest obstacles to social and economic development. Whereas there has been extensive research identifying the causes of corruption, there is little experimental research on the impact of political institutions on corruption using designs that control for significant confounders. This paper uses a series of laboratory experiments conducted in 2013 Egypt in which a government official decides whether to spend tax revenues paid by subjects on a self-serving good or a good that benefits everyone equally. We have two experimental manipulations: (a) whether the official is electorally accountable to subjects or not; (b) whether subjects could send messages of protests to the official (and one another). We find evidence that electoral accountability does decrease the probability of the official choosing the self-serving good by 17% whereas voice accountability generates such outcome only in the authoritarian treatment (a reduction of corruption by 29%). We also find suggestive evidence that, in the authoritarian treatment, the likelihood of funding the self-serving good decreases by 27% when taxes paid by citizens fall short of the official’s threshold. Our contribution to the literature is two-fold: (a) we are able to single out the effect of specific democratic mechanisms on government corruption; (b) we test outcomes of democratic mechanisms on a traditionally understudied subject pool.