The assumption that optimization results depend on coordinate system selected to describe a biomechanical model is tested by comparing two solutions obtained with generalized and natural coordinate systems. A 5-degrees of freedom planar musculoskeletal model actuated by 9 Hill-type musculotendon units was created to simulate lifting a leg up. Each individual muscle force was discretized into a set of independent design variables, and an inverse dynamic parameter optimization method was used in the computations. The optimal time characteristics of the predicted muscle forces for both solutions are presented. Some remarks concerning the efficiency of natural coordinates for solving optimal control problems are also included.