We investigate the hypothesis that the direction of the propulsion force in manual wheelchair propulsion can be interpreted as a result of the balance between the mechanical task requirements and the driver's biomechanical possibilities. We quantify the balance at the joint level in the form of an effect-cost criterion, from which we predict the force direction that results in an optimal compromise. Kinematic and dynamic data were collected from nine habitual wheelchair users driving at four velocities (0.83, 1.11, 1.39, 1.67 m/s) and three external power levels (10, 20, 30 W). Experimental data and predictions are in good agreement in the middle and final part of the push; the effect-cost value in this region approximates the achievable maximum. Early in the push the effect-cost criterion predicts an upwards propulsion force whereas the experimental force is downwards, the difference probably being mainly attributable to the force generation dynamics of the muscles. As a result of the geometric features of large-rim manual wheelchairs, the mechanically required and biomechanically preferred force directions are not in accordance during a substantial part of the push, making even the best compromise a poor one. This may contribute to the low mechanical efficiency of manual wheelchair propulsion and the high incidence of shoulder complaints.