Biomechanical models which require information on, e.g., joint torque and muscle force are useful in the estimation of when and how mechanical overload of the musculoskeletal system may lead to disorders. The aim was to study the reliability and validity of magnetic resonance imaging (MRI) to quantify muscle sizes and moment arms by MRI and to test selected anthropometric measures as predictors of muscle sizes and moment arms. A total of 20 healthy Scandinavian women (age 22–58 years) participated in an MRI scanning of their dominant shoulder. With a PC-based program the reliability and the validity of the MRI measurements was estimated to be high, and mean anatomical cross-sectional areas (ACSA) and muscle lengths were measured to be 4.0, 9.8 and 12.1 cm² and 12.0, 12.6 and 12.8 cm for m. supraspinatus, m. infraspinatus and m. subscapularis, respectively. Volumes were calculated to be 48.8, 125.1 and 153.6 cm³. Moment arms were measured with the upper arm in a neutral position and in a functional position of 34° abduction for m. supraspinatus only, and were 2.4 and 2.6 cm. Physiological cross-sectional area (PCSA) and its fiber force component were estimated from dissected fiber length and pennation angle. MRI volume and PCSA were 1.4–1.7 times higher than dissection data, primarily because of age differences. No external anthropometric measures were found to be predictors of volumes or moment arms.