Musculoskeletal models and optimization methods are combined to calculate muscle forces. Some model parameters cannot be experimentally measured due to the invasiveness, such as the muscle moment arms or the muscle and tendon lengths. Moreover, other parameters used in the optimization, such as the muscle synergy components, can be also unknown. The estimation of all these parameters needs to be validated to obtain physiologically consistent results. In this study, a two-step optimization problem was formulated to predict both muscle and knee contact forces of a subject wearing an instrumented knee prosthesis. In the outer level, muscle parameters were calibrated, whereas in the inner level, muscle activations were predicted. Two approaches are presented. In Approach A, contact forces were used when calibrating the parameters, whereas in Approach B, no contact force information was used as input. The optimization formulation is validated comparing the model and the experimental knee contact forces. The goal was to evaluate whether we can predict the contact forces when in-vivo contact forces are not available.