We estimate habit formation in voting—the effect of past on current turnout—by exploit-ing transitory voting cost shocks. Using county-level data on U.S. presidential elections from 1952-2012, we find that rainfall on current and past election days reduces voter turnout. Our estimates imply that a 1-point decrease in past turnout lowers current turnout by 0.6-1.0 points. Further analyses suggest that habit formation operates by re-inforcing the direct consumption value of voting and that our estimates may be amplified by social spillovers.