The California class size reduction program provided schools with cash rewards for K-3 classes of 20 or fewer students. I show how program rules made it possible for schools to save money by using mixed-grade classes to meet class size reduction obligations while maintaining larger average class sizes. I also show that this smoothing of students across grades is associated with a significant test score gap for certain second and third grade students. My estimates suggest that class size reduction may lead to lower test scores for students placed in combination classes. Alternative explanations of teacher experience and credentialing changes cannot explain the test score pattern. This result spotlights both the underappreciated role of age heterogeneity in classroom learning and the difficulty of replicating the success of policy experiments in statewide reform. (Contains 6 tables, 5 figures and 20 footnotes.)