How do violence and the subsequent repression impact electoral results in non-consolidated democratic contexts? While some argue that violence and/or repression quell political collective action, others sustain that it can backfire and stimulate political mobilization--the repression-mobilization puzzle. Yet, our understanding of the effect of an episode of riots and repression on electoral behaviour is still limited, especially in the context of a non-fully democratic system. This article examines an episode of riots and violence and whether, and in what direction, the event impacted parties' share of the votes. We focus on a type of violence that was neither perpetrated by the government nor directly directed towards the government: In particular, we exploit an episode of large-scale riots and violence that took place in Barcelona in 1909, where the working class protested against conscription by burning a large number of religious buildings. However, neither all churches were burnt nor the intensity of violence was equal all over the city. Our empirics rely on a new dataset built using fine-grained geolocated information, including the location of protests and repression, and different indicators related to electoral behaviour. Using a difference-in-differences framework, we assess whether this violence further mobilized left-wing supporters or, in turn, made part of the middle and mid-low classes switch their support towards a law and order political party.