Racial disparities in policing are well documented. Two standard explanations are police officer prejudice and unequal crime rates across communities. We analyze a behavioral theory of policing that incorporates both, while also relaxing the standard (but unrealistic) assumption that decision makers must be fully Bayesian. We analyze two versions of the theory. Street-level behavioral policing features an officer who mistakenly conflates experiences of crime on the job with those in his personal life. Institutionalized behavioral policing features a police chief who makes assessments about city-wide crime rates using crime statistics that do not account for differential policing intensities. Strikingly, our analysis demonstrates that discrimination can occur without either prejudice or unequal crime rates. In both cases, disparate treatment and inaccurate beliefs about groups’ relative criminality are joint and mutually reinforcing outcomes. Moreover, behavioral policing can exacerbate discrimination even when police officials remain convinced they don’t discriminate.