Background: Efficiency in manual wheelchair propulsion is low, as is the fraction of the propulsion force that is attributed to the moment of propulsion of the wheelchair. In this study we tested the hypothesis that a tangential propulsion force direction leads to an increase in physiological cost, due to (1) the sub-optimal use of elbow flexors and extensors, and/or (2) the necessity of preventing of glenohumeral subluxation. Methods: Five able-bodied and 11 individuals with a spinal cord injury propelled a wheelchair while kinematics and kinetics were collected. The results were used to perform inverse dynamical simulations with input of (1) the experimentally obtained propulsion force, and (2) only the tangential component of that force. Findings: In the tangential force condition the physiological cost was over 30% higher, while the tangential propulsion force was only 75% of the total experimental force. According to model estimations, the tangential force condition led to more co-contraction around the elbow, and a higher power production around the shoulder joint. The tangential propulsion force led to a significant, but small 4% increase in necessity for the model to compensate for glenohumeral subluxation, which indicates that this is not a likely cause of the decrease in efficiency. Interpretation: The present findings support the hypothesis that the observed force direction in wheelchair propulsion is a compromise between efficiency and the constraints imposed by the wheelchair-user system. This implies that training should not be aimed at optimization of the propulsion force, because this may be less efficient and more straining for the musculoskeletal system. © 2008 Elsevier Ltd. All rights reserved.