Elections in developing countries commonly fail to deliver accountability because of manipulation, often involving collusion between corrupt election officials and political candidates. We report the results of an experimental evaluation of Quick Count Photo Capture—a monitoring technology designed to detect the illegal sale of votes by corrupt election officials to candidates—carried out in 471 polling centers across Afghanistan during the 2010 parliamentary elections. The intervention reduced vote counts by 25% for the candidate most likely to be buying votes and reduced the stealing of election materials by about 60%. Additionally, we investigate the role of corrupt institutions in facilitating election fraud by combining: (i) separate fraud measurements at three important stages of the election; (ii) rich data on the political connections of key parliamentary candidates; (iii) precise geographic coordinates of polling centers; and (iv) experimental variation from our evaluation. Interestingly, strong political candidates react to the intervention by substituting fraud spatially and weak candidates react by substituting temporally. We explain these results in the context of a theory of corrupt vote transactions in which the capacity of candidates to protect corrupt officials from prosecution determines equilibrium levels of spatial and temporal substitution. Video with author: Hear Michael Callen, discuss how smart phones may save elections.