How much of the geographic clustering of economic activity is attributable to agglomeration spillovers as opposed to natural advantages? I present evidence on this question using data on the long-run effects of large scale hydroelectric dams built in the U.S. over the 20th century, obtained through a unique comparison between counties with or without dams but with similar hydropower potential. Until mid-century, the availability of cheap local power from hydroelectric dams conveyed an important advantage that attracted industry and population. By the 1950s, however, these advantages were attenuated by improvements in the efficiency of thermal power generation and the advent of high tension transmission lines. Using a novel combination of synthetic control methods and event-study techniques, I show that, on average, dams built before 1950 had substantial short run effects on local population and employment growth, whereas those built after 1950 had no such effects. Moreover, the impact of pre-1950 dams persisted and continued to grow after the advantages of cheap local hydroelectricity were attenuated, suggesting the presence of important agglomeration spillovers. Over a 50 year horizon, I estimate that at least one half of the long run effect of pre-1950 dams is due to spillovers. The estimated short and long run effects are highly robust to alternative procedures for selecting synthetic controls, to controls for confounding factors such as proximity to transportation networks, and to alternative sample restrictions, such as dropping dams built by the Tennessee Valley Authority or removing control counties with environmental regulations. I also find small local agglomeration effects from smaller dam projects, and small spillovers to nearby locations from large dams.