Violence in conflict settings is seldom random, making its effects indistinguishable from the intentions of the perpetrator. We leverage on the quasi-randomness of accidental landmine explosions to study how violence shapes electoral outcomes in Colombia. We combine the geolocation of landmine blasts with the coordinates of voting polls in a regression discontinuity design that compares polls close to which a landmine exploded just before the election to those close to which it did just afterward. Blasts within a month from election day depress turnout by 23%. In addition, those who do vote penalize the democratic left for the explosions and are more likely to support political parties with ties with illegal paramilitary groups.