Thigh muscle volume was assessed using magnetic resonance imaging in 16 subjects with spinal muscular atrophy. Scans were successful for 14 of 16 subjects (1 type 1, 6 type 2, and 7 type 3) as young as 5.7 years. Muscle volume with normal and abnormal signal was measured using blinded, semiautomated analysis of reconstructed data. Results were compared with segmental lean mass estimated by dual-energy X-ray absorptiometry and correlated with clinical and electrophysiological measures of disease severity. Muscle volume was reduced with abnormal signal quality. Test-retest reliability (r = .99) and correlation with dual-energy X-ray absorptiometry (r = .91) were excellent. Type 2 subjects had lower volume (3.5 ± 1.6 vs 6.3 ± 2.8 mL/cm height; P = .06) and higher percentage of muscle with abnormal signal (68% ± 20% vs 47% ± 27%; P = .14) than type 3. Reproducibility, tolerability, and strong correlation with clinical measures make magnetic resonance imaging a candidate biomarker for clinical research.