State interventions against organized criminal groups (OCGs) sometimes work to improve security, but often exacerbate violence. To understand why, this article offers a theory about criminal governance in five types of criminal regimes—Insurgent, Bandit, Symbiotic, Predatory, and Split. These differ according to whether criminal groups confront or collude with state actors, abuse or cooperate with the community, and hold a monopoly or contest territory with rival OCGs. Police interventions in these criminal regimes pose different challenges and are associated with markedly different local security outcomes. We provide evidence of this theory by using a multimethod research design combining quasi-experimental statistical analyses, automated text analysis, extensive qualitative research, and a large-N survey in the context of Rio de Janeiro’s “Pacifying Police Units” (UPPs), which sought to reclaim control of the favelas from criminal organizations.