Previous correlational and quasi-experimental studies of arts integration--the pedagogical practice of "teaching through the arts"--suggest its value for enhancing cognitive, academic, and social skills. This study reports the results of a small, preliminary classroom-based experiment that tested effects of arts integration on long-term retention of content. We designed matched arts-integrated (AI) and conventional science units in astronomy and ecology. Four randomized groups of 5th graders in one school completed one unit in the treatment (AI) condition and the other in the control (conventional) condition. To control for teacher effects, four teachers taught the same subject to different groups in each condition. We administered curriculum-based assessments before, immediately after, and 2 months after each unit to measure initial learning and retention. Results showed no differences in initial learning, but significantly better retention in the AI condition. Increases in retention were greatest for students at the lowest levels of reading achievement.