State-wide reports on police traffic stops and searches summarize very large populations, making them potentially powerful tools for identifying racial bias, particularly when statistics on search outcomes are included. But when the reported statistics conflate searches involving different levels of police discretion, standard tests for racial bias are not applicable. This article develops a model of police search decisions that allows for nondiscretionary searches and derives tests for racial bias in data that mix different search types. Our tests reject unbiased policing as an explanation of the disparate impact of motor-vehicle searches on minorities in Missouri.