Methods for developing mathematical models representing entire human muscles are briefly reviewed, with special emphasis on aspects of modelling velocity dependence using cross-bridge dynamics, and isometric force–length properties from myofilament lengths and muscle architecture. For each of these components, mechanistic (using basic contraction mechanisms) and phenomenological (“black-box”) models are available. Experiments on constant-velocity lengthening at different velocities were simulated using (a) a cross-bridge based model and (b) a Hill-based model. The Hill model was superior in its ability to predict muscle forces under different conditions with the same model parameters. Regarding force–length properties, myofilament overlap and muscle architecture did not correctly predict maximal isometric joint moments over the entire functional range of motion. The width of the force–length relationship of all contractile elements in a lower extremity model may be optimized to fit measured isometric moment–angle relationships. The resulting increase in width suggests that for some short-fibered muscles with complex architecture, the “effective” muscle fibre length is increased because muscle fibres may be partly connected in series as well as in parallel. It is concluded that a hybrid phenomenological/mechanistic muscle model is most likely to be practical (i.e. parameters can be estimated for human muscle) as well as accurate (i.e. correct forces are predicted for a wide range of conditions).