The validity of current inverse dynamics models utilized for motion analysis is investigated. It is shown that observables generated by the real biosystem, such as ground reaction forces, are incompatible with comparable responses of skeletodynamical inverse models currently in use. This implies that results obtained with such models are erroneous to varying degrees while a quantification of these errors is difficult or impossible. This phenomenon is termed the fundamental myoskeletal inverse dynamics problem. A model fidelity indicator is proposed which, for a specific inverse dynamics model applied to a particular motion, provides a dimensionless numerical measure for the replicative validity of that model and the fidelity of its input data. A practical example demonstrates the usefulness of this indicator. It is suggested that the development of structurally sufficiently complex and biologically more realistic skeletomechanical models as well as substantial error reductions in data measuring and processing procedures will be necessary to improve the accuracy of inverse dynamics model computations.