Despite an abundance of empirical evidence on crime spanning over 40 years, there exists no consensus on the impact of the criminal justice system on crime activity. We construct a new panel data set that contains all relevant variables prescribed by economic theory. Our identification strategy allows for a feedback relationship between crime and deterrence variables, and it controls for omitted variables and measurement error. We deviate from the majority of the literature in that we specify a dynamic model, which captures the essential feature of habit formation and persistence in aggregate behaviour. Our results show that the criminal justice system exerts a large influence on crime activity. Increasing the risk of apprehension and conviction is more influential in reducing crime than raising the expected severity of punishment.