Decentralization can improve service delivery, but it can also generate externalities across jurisdictional boundaries. We examine the nature and size of water pollution externalities as rivers flow across jurisdictions. Panel data on water pollution in Brazilian rivers coupled with county splits that change the locations of borders allow us to identify the spatial patterns of pollution as rivers approach and cross borders, controlling for fixed effects and trends specific to each location. The theory of externalities predicts that pollution should increase at an increasing rate as the river approaches the downstream exit border, that there should be a structural break in the slope of the pollution function at the border, and that a larger number of managing jurisdictions should exacerbate pollution externalities. We find support for all four predictions in the data. The results are robust to a number of tests designed to address endogeneity concerns about the location of county splits. We also show that it is difficult to reconcile other theories and mechanisms with the four pieces of congruent evidence we report. Satellite data on growth in night-time lights along rivers shows that local authorities allow more settlements to develop close to rivers in the downstream portions of counties, which is the likely underlying mechanism. The border effects on pollution are not as pronounced when the cost of inter-jurisdictional coordination is lower (after a water basin committee forms, or when political affiliations of mayors in upstream and downstream counties match).