Looking for new opportunities in mechanical design, we are interested in studying the kinematic behaviour of biological joints. The real kinematic behaviour of the elbow of quadruped animals (which is submitted to high mechanical stresses in comparison with bipeds) remains unexplored. The sheep elbow joint was chosen because of its similarity with a revolute joint. The main objective of this study is to estimate the effects of elbow simplifications on the prediction of joint reaction forces in inverse dynamic simulations. Rigid motions between humerus and radius-ulna were registered during full flexion-extension gestures on five cadaveric specimens. The experiments were initially conducted with fresh specimens with ligaments and repeated after removal of all soft tissue, including cartilage. A digital image correlation system was used for tracking optical markers fixed on the bones. The geometry of the specimens was digitized using a 3D optical scanner. Then, the instantaneous helical axis of the joint was computed for each acquisition time. Finally, an OpenSim musculoskeletal model of the sheep forelimb was used to quantify effects of elbow joint approximations on the prediction of joint reaction forces. The motion analysis showed that only the medial-lateral translation is sufficiently large regarding the measuring uncertainty of the experiments. This translation assimilates the sheep elbow to a screw joint instead of a revolute joint. In comparison with fresh specimens, the experiments conducted with dry bone specimens (bones without soft tissue) provided different kinematic behaviour. From the results of our inverse dynamic simulations, it was noticed that the inclusion of the medial-lateral translation to the model made up with the mean flexion axis does not affect the predicted joint reaction forces. A geometrical difference between the axis of the best fitting cylinder and the mean flexion axis (derived from the motion analysis) of fresh specimens was highlighted. This geometrical difference impacts slightly the prediction of joint reactions.