Is it feasible in the current United States to administer voter identification laws in a race-neutral manner? We study this question using rigorous field methods and state-of-the-art statistical techniques, thus accounting for sources of uncertainty (including survey non-response and clustering) that previous studies ignore. We conduct a sensitivity analysis to account for voters who were legally required to have been asked for ID under federal and state law. We conduct an experiment with a training program that clarified proper ID law administration. Finally, we study a jurisdiction and an election in which administration of ID laws was unlikely to pose issues of racial difference, and in which (under the law) the decision to request an ID was nondiscretionary. We find strong evidence that Hispanic and black voters were asked for identification at higher rates than white voters, even after adjusting for a number of other factors. The magnitudes of the differences are significant. We explore the theoretical and legal consequences of our findings.