We study the effects of broken windows policing on crime using geo-located crime and arrest reports for 80 Colombian cities. Broadly defined, broken windows policing consists of intensifying arrests - sometimes for minor offenses - to deter potential criminals. To estimate causal effects, we build grids of 200 âˆšÃ³ 200 meters over the urban perimeter of all cities and produce event studies to look at the effects of shocks in police activity in the periods to follow. We use spikes in the number of arrests with no warrant - which are more likely associated with unplanned police presence - as a proxy for shocks in broken windows policing. As expected, we observe an increase in crimes during the shock period, as each arrest implies at least one crime report. In the following periods, crimes decrease both in the place of the arrests and the surroundings. With many treated grids and many places exposed to spillovers, these effects add up. On aggregate, the crime reduction offsets the observed increase during the shock period. Direct effects are more immediate and precise at low crime grids, but beneficial spillovers seem more relevant at crime hot spots. The effects of broken windows policing circumscribe to cities with low or moderate organized crime, consistent with criminal organizations planning their activities more systematically than disorganized criminals.