We present results from laboratory experiments studying statistical discrimination and affirmative action. We induce statistical discrimination in simple labor-market interactions between firms and workers. We then introduce affirmative-action policies that vary in the size and duration of a subsidy that firms receive for hiring discriminated against workers. These different affirmative-action policies have nearly the same effect, and practically eliminate discriminatory hiring practices. However, once lifted, few positive effects remain and discrimination reverts to its initial levels. One exception is lengthy affirmative-action policies, which exhibit somewhat longer-lived effects. Stickiness of beliefs, which we elicit, helps explain the observed outcomes.