The modelling of skeletal muscle raises a number of philosophical questions, particularly in the realm of the relationship between different possible levels of representation and explanation. After a brief incursion into this area, a list of desiderata is proposed as a guiding principle for the construction of a viable model, including: comprehensiveness, soundness, experimental consistency, predictive ability and refinability. Each of these principles is illustrated by means of simple examples. The presence of internal constraints, such as incompressibility, may lead to counterintuitive results. A one-panel example is exploited to advocate the use of the principle of virtual work as the ideal tool to deal with these situations. The question of stability in the descending limb of the force–length relation is addressed and a purely mechanical analogue is suggested. New experimental results confirm the assumption that fibre stiffness is positive even in the descending limb. The indeterminacy of the force–sharing problem is traditionally resolved by optimizing a, presumably, physically meaningful target function. After presenting some new results in this area, based on a separation theorem, it is suggested that a more fundamental approach to the problem is the abandoning of optimization criteria in favour of an explicit implementation of activation criteria.