The authors present a simple classroom game in which students are randomly designated as employers, purple workers, or green workers. This environment may generate "statistical" discrimination if workers of one color tend not to invest because they anticipate lower opportunities in the labor market, and these beliefs are self-confirming as employers learn that it is, on average, less profitable to hire workers of that color. Such discriminatory equilibria may arise even when workers are ex-ante identical, and the employer has no prior information regarding potential workers. The exercise typically generates a lively discussion about discrimination and how it may be addressed by alternative public policies. (Contains 5 endnotes.)