Instrumental accounts of election violence suggest that politicians use violence to reduce turnout among non-supporters. Departing from the conventional wisdom, we argue that elites use violence not only for its deterrent but also for its mobilizing effects. Our paper establishes two instrumental logics: First, violence is used to scare non-supporters and deter them from voting, as current work implies. Second, violence is also employed to polarize the electorate and increase support for politicians complicit with or responsible for the violence. We argue that the two instrumental logics vary in terms of actors, geographical targeting, cleavages of contention, and empirical implications for non-violence outcomes. We expect that violence for demobilization is used by incumbents to deter non-supporters in opposition strongholds, results in lower electoral turnout, and is carried out along partisan lines. In contrast, we argue that violence for polarization is used by challenger parties to mobilize new supporters. We expect such violence in electorally competitive areas, that violence results in higher electoral turnout for the challenger, and that it exploits non-partisan cleavages, such as social, cultural, or religious divisions. Empirically, we assess our expectations using a novel micro-level data for more than 90,000 polling booths from West Bengal in India—a context that matches our scope conditions of routine political violence and competitive elections.