The recent protests regarding the state of policing in the United States clearly demonstrate that how the police do their job creates a salient potential for harm to the public. This study applies a multidimensional paradigm of risk perception to quantify evaluations of police-caused harm. Using data from a national (U.S.) convenience sample (n = 1,890) that oversampled individuals who self-identified as black or Muslim, we tested whether these evaluations vary systematically (using confidence intervals), whether they covary with police legitimacy (using structural equation modeling), and the extent to which that covariance differs by demographic status (using multiple groups structural equation modeling). Our results suggest that black and Muslim individuals evaluate police-caused harm differently than do majority group members (white and Christian) on most, but not all, of the measured dimensions. We also find that those evaluations are predictive of trust and provide evidence of some level of consistency across communities.