Social promotion has long been the normal practice in American schools. Critics of this practice, whereby students are promoted to the next grade regardless of academic preparation, have suggested that students would benefit academically if they were made to repeat a grade. Supporters of social promotion claim that retaining students disrupts them socially, producing greater academic harm than promotion would. A number of states and school districts have attempted to curtail social promotion, by requiring students to demonstrate academic preparation on a standardized test before they can be promoted to the next grade. This study analyzes the effects of Florida's test-based promotion policy on student achievement two years after initial retention. It builds upon a previous evaluation of the policy in two ways. First, the authors examine whether the initial benefits of retention observed in the previous study continue, expand, or contract in the second year after students are retained. Second, the researchers determine whether discrepancies between their evaluation and the evaluation of a test-based promotion policy in Chicago are caused by differences in how researchers examined the issue, or by differences in the nature of the programs. Analysis shows that, after two years of the policy, retained Florida students made significant reading gains relative to the control group of socially promoted students. These academic benefits grew substantially from the first to the second year after retention. That is, students lacking in basic skills who are socially promoted appear to fall farther behind over time, whereas retained students appear to be able to catch up on the skills they are lacking. Further, these positive results in Florida were found both when we use the same research design that used in the previous study and when using a design similar to that employed by the evaluation of the program in Chicago. The differences between the Chicago and Florida evaluationsappear to be caused by differences in the details of the programs, and not by differences in how the programs were evaluated. (Contains 10 endnotes and 5 tables.)