This dissertation investigated the ability of optimization-based biomechanical models to predict torso muscular activity. Two optimization models were considered: the Minimum Intensity Compression (MIC) model and the Sum of the Cubed Intensities (SCI) model. For each model, two sets of muscle geometries (moment arms, lines of action, and cross-sectional areas) were used as inputs: one was a compilation of several studies and one was reported by Han, Ahn, Goel, Takeuchi, and McGowan (1992). For each of the four model combinations, either 10 or 18 muscles were used to formulate predictions. With computer simulations, the four models were used to predict muscle. forces under loading conditions including three types of moments, flexion/extension, lateral bending, and torsional. The results indicated large differences in the predicted forces due to the different models and muscle geometries. Changes in force predictions for identical muscles were also found when the number of muscles in the formulation increased from 10 to 18. The models also predicted varying active and inactive regions of the muscles in response to changing moments. To determine empirically the activity of muscles and test the accuracy of optimization-based models and inputs, combinations of the three moments were also applied to subjects through loads held in the hands. The subjects maintained a static posture during physical exertions while attempting flexion/extension, lateral bending, and torsion to counter the external loads. Results indicated little improvement in prediction of actual muscle activity by including 18 muscles instead of 10. The SCI model with compilation geometry provided the best predictions of actual muscle activity judged by overall R? values for each muscle and by the number of subjects the model accounted for over 50% of the variation. Actual activities of five of the eight muscles monitored were well predicted: left and right rectus abdominis, left external oblique, and left and right erector spinae. The left and right latissimus dorsi were poorly predicted by the models, which was due to the use of the muscle in shoulder Stability which was not accounted for in trunk optimization models. The experimental method included application of moments to subjects through loads held by the hands and by loads attached to a harness mounted at the shoulder level. The left and right erector spinae were the only muscles which exhibited the same activity for both apparatuses used to apply moments. The left and right latissimus dorsi showed the largest increase in activity when loads were held with the hands instead of applied via the harness. Differences between hand and harness loading were also found for the left and right rectus abdominis muscles and left and right external obliques at varying moment conditions. / Ph. D.