Most discussion of special education has centered on the costs of providing mandated programs for children with disabilities and not on their effectiveness. As in many other policy areas, inferring program effectiveness is difficult because students not in special education do not provide a good comparison group. By following students who move in who move in and out of targeted programs, however, we are able to identify program effectiveness from changes over time in individual performance. we find that the average special education program significantly boosts mathematics achievement of special-education students, particularly those classified as learning-disabled, or emotionally disturbed, while not detracting from regular-education students. These results are estimated quite precisely from models of students and school-by-grade-by-year fixed effects in achievement gains and they are robust to a series of specification tests.