Nineteen percent of 1997-98 North Carolina 3rd graders were observed to drop out of high school. A series of logits predict probabilities of dropping out on determinants such as math and reading test scores, absenteeism, suspension, and retention, at the following grade levels: 3rd, 5th, 8th, and 9th. The same cohort and variables are used to estimate benefits to the 15,737 students admitted to a special program ostensibly for academically and intellectually gifted children. I estimate the probability of admission for schoolmates with similar ability in math and reading to be substantially higher for those from upper income households. Finally, I conclude that extending similar resources to an equal number of high-risk students, as determined by their 3rd grade predicted probabilities, would lead to a 25% reduction in the total cohort dropout rate, and that even dividing existing resources between the two groups could cut dropout rates by half that. (Contains 6 tables and 5 figures.)