El Salvador, Honduras and Guatemala count among today's most violent countries of the world. Qualitative research has claimed that large-scale deportations of Central American convicts have played an important role for the spread of gangs and rampant violence in the region. Using a novel identification strategy, this paper provides the first econometric evidence for this hypothesis from the case of El Salvador. Regarding the dependent variable, the policy experiment of a truce between rivaling gangs in 2012 allows to single out gang-related killings from overall homicide rates. The explanatory variable exploits subnational variation in the exposure of migrant communities to exogenous conditions in the host country. Violence spilled over to migrants' places of origin when migrant corridors developed around US destinations with high pre-existing levels of violent crime. The cross-sectional evidence is backed by panel data analysis dating back to 1999. The annual inflow of convicts translated into rising homicides mainly in those municipalities whose migrants were exposed to high pre-existing crime at destination, whereas deportations of non-convicts did not have the same effect. These finding are in line with evidence on the origin of Central American gangs in US cities and convicts' return to their places of birth after massive deportations since the mid-1990s.