Optimization-based models for prediction of muscle forces in the lumbar region of the torso are used to estimate the forces acting on spinal motion segments, especially for asymmetric tasks. The objectives of this study were to determine (a) which of four torso model formulations best predicted the electromyographic data, (b) the difference in muscular contribution to spinal compression force for the four models, and (c) the effect of using the lowest possible muscle stress bound in the model formulation. An approach for the investigation of competing optimization model formulations was developed and was illustrated with electromyographic data from static asymmetric loading conditions. This method is based on (a) the choice of experimental conditions in which models predict decidedly different muscle forces, and (b) the ability to ensure that the experimental conditions are such that the minimum number of assumptions about the force-electromyogram relationship must be made in order to choose between competing model predictions. Of the four models analyzed, only the formulation with an objective function that was the sum of cubed muscle stresses predicted the electromyographic data acceptably. The muscular contribution to spinal compression force predicted by these models differed by as much as 160% for some experimental conditions. The use of the lowest possible muscle stress bound does not appear to predict muscle forces that are in agreement with electromyographic data.