Upward Bound (UB) was one of the original federal "Great Society" programs of the 1960s and remains, fifty years later, the single largest college access program in the country. Recently, Congress has reduced funding and considered eliminating the program because of federal budget pressures and because the first analysis of the only national randomized trial concluded that UB had "no detectable effect." In this study, we explain problems with the study design that have made this conclusion controversial and made it difficult to identify average treatment effects that are unbiased for the program population. The study design is sufficient, however, to test other important hypotheses. We show that UB has drifted away from its original intent of serving disadvantaged students and become less efficient in the process. Specifically, over time, UB has come to serve students whose parents have higher incomes and education levels. Also, program administrators deem students ineligible based on misbehavior and other factors, even though assignment to the UB treatment increases this ineligible group's probability of graduating high school and obtaining some type of college credential by 6-10 percentage points. We show that the program is cost-effective for this group and would reduce achievement gaps if it were better targeted. In short, the government should encourage sites to get back to the program roots of serving more disadvantaged students. An appendix includes discussion of key variables, provides additional formal analysis of the influence of weight errors in effect heterogeneity analysis, and includes additional tables and figures.