Cognitive skills are associated with academic performance, but little is known about how to improve these skills in the classroom. Here, we present the results of a pilot study in which teachers were trained to engage students in cognitive skill practice through playing games. Fifth-grade students at an experimental charter school were randomly assigned to receive cognitive skill instruction (CSI) or instruction in geography and typing. Students in the CSI group improved significantly more than the control group on a composite measure of cognitive skills. CSI was more effective for students with lower standardized test scores. Although there was no group effect on test scores, cognitive improvement correlated positively with test score improvement only in the CSI group. Beyond showing that cognitive skills can be improved in the classroom, this study provides lessons for the future of CSI, including changes to professional development and challenges for scalability.