African Americans face shorter employment durations than similar whites. We hypothesize that employers discriminate in acquiring or acting on ability-relevant information. In our model, monitoring black but not white workers is self-sustaining: new black hires are more likely to have been screened by a previous employer, causing firms to discriminate in monitoring. We confirm our predictions that the layoff hazard is initially higher for black workers but converges to that for white workers, and declines with AFQT more sharply for black workers. Two additional predictions, lower lifetime incomes and longer unemployment durations for black workers, are known to be empirically supported.