Mimicking the so-called great American crime decline, violent crime in the state of São Paulo dropped sharply in the 2000s after rising steadily throughout the 1980s and 1990s. This paper evaluates the role of crack cocaine in explaining the aggregate dynamics in violence. Four facts are established. First, the aggregate data show a tight comovement between the prevalence of crack cocaine and homicides. Second, using city-level apprehension and possession data, I find a strong elasticity of violent crime with respect to crack cocaine after controlling for year fixed effects, city effects, and many time-varying covariates. I use the estimated elasticity to compute the contribution of crack cocaine to aggregate violence. Crack explains 30 percent of time series variation in the data. Third, only drug trafficking—not drug possession—has an impact on homicides. Finally, I find no impact on property crimes, a weaker impact on attempted murder, and, interestingly, a weak negative impact on aggravated assault. The theory suggests that both facts—only trafficking matters and crack affects only homicides, not property crime—can be rationalized only if drug-induced crime is driven by systemic violence induced by illegality itself. These results are important for policy because they suggest that violence will not follow the legalization of both the possession and trade of powdered cocaine or crack cocaine.