This paper examines how assumptions imposed on the data influence estimates of schooling's effect on earnings. The paper models schooling decisions as treatment effects and imposes assumptions about schooling selection to estimate bounds on the treatment effect. The study begins by using the worst-case bounds derived by Manski (1989, 1990, 1994, 1995) and adds assumptions from the Roy model of schooling self-selection to narrow the bounds on the schooling treatment effect. The bounds are narrowed further by using family structure, college proximity, and school-quality characteristics as exclusion restrictions. The selection problem requires the researcher to make explicit assumptions to estimate the effect of schooling on earnings. This paper demonstrates that different selection assumptions yield very different results.