This study estimates the causal effect of homeless shelters on neighborhood crime. Using a novel dataset of geocoded homeless housing sites in the City of Los Angeles and a spatial difference-in-differences design that exploits the timing of shelter openings and spatial variation in exposure, I estimate that the opening of a homeless shelter increases overall crime in its immediate vicinity, with little evidence of effects beyond the surrounding neighborhood. Suggestive evidence indicates that this effect may be driven by an increase in the local concentration of homeless individuals, while significant effects are found for crimes involving homeless victims but not homeless suspects. These findings suggest that negative externalities of homeless shelters are highly localized and have important implications for the siting and design of homeless housing policies.