We conduct a study under 2,400 third grade students at three large secondary comprehensive schools to evaluate a gifted and talented (GT) program with selective program admission based on past achievement. We construct three complementary estimates of the program's impact on student achievement. First, we use the fragmented GT program implementation (in different tracks at different schools) to get difference-in-differences (DD) estimates for all students above the admission cutoff. Second, we use the GT admission rule to get regression discontinuity (RD) estimates for students near the admission cutoff. And third, we combine the DD and RD designs to estimate how the program's impact varies with past achievement. We find that all participating students do better because of the GT program. Students near the admission cutoff experience a 0.2 standard deviation gain in their grade point average. Students further away from the admission cutoff experience larger gains.