Abstract -We estimate the impact of fire-armed police on violent crimes (homicides and acts of aggression) in a quasi-natural experiment in Brazil. In 2003, Brazilian legislators approved a law that regulates the use of firearms by the municipal police. It establishes that municipalities with 50 thousand or more inhabitants are allowed to have local police with firearms. We explore this population eligibility criterion as an instrumental variable of firearm possession in a linear regression discontinuity design. Using different data sources at the municipality level for selected years from 2002 to 2012, we find robust results that a municipal police force with firearms significantly reduces crime compared with a municipal police force without firearms. Moreover, we find suggestive evidence that the effect is partly due to greater police efforts such as more arrests and the incapacitation of criminals. Finally, we demonstrate that one of possible effect (spillover) of deterrence (i.e., migration of potential criminals from one municipality to its neighbor) is not detected.