Criminal law enforcement depends on the actions of public agents such as police officers, but the resulting agency problems have been neglected in the law and economics literature (especially outside the specific context of corruption). We develop an agency model of police behavior that emphasizes intrinsic motivation and self-selection. Drawing on experimental evidence on punishment preferences, in which subjects reveal a heterogeneous preference for punishing wrongdoers, our model identifies circumstances in which “punitive” individuals (with stronger-than-average punishment preferences) will self-select into law enforcement jobs that offer the opportunity to punish (or facilitate the punishment of) wrongdoers. Such “punitive” agents will accept a lower salary, but create agency costs associated with their excessive zeal (relative to the public’s preferences) in searching, seizing, and punishing suspects. In our framework, the public chooses (under reasonable assumptions) to hire punitive police agents, while providing suspects with strong criminal procedure protections, thereby empowering other agents (such as the judiciary) with average punishment preferences to limit the agency costs of excessive zeal. We thus argue that intrinsic motivation and self-selection provide a possible explanation for the bifurcated structure of criminal law enforcement in which courts constrain police with pro-defendant rules of criminal procedure. We also explore various other implications of this framework.