We show that societies with a history of irrigation-based agriculture have been less likely to adopt democracy than societies with a history of rainfed agriculture. Rather than actual irrigation, the empirical analysis is based on how much irrigation potentially can increase yields.Irrigation potential is derived from a range of exogenous geographic factors, and reverse causality is therefore ruled out. Our results hold both at the cross-country level, and at the subnational
level in premodern societies surveyed by ethnographers.