Soft tissue artefacts (STA) distort marker-based knee kinematics. Global optimization (GO) compensates for STA, but it lacks of direct validation on subjects. This study evaluates the performance of eight multi-joint models used in GO. Ten subjects were recruited: five healthy and five osteoarthritis subjects. Each subject performed dynamic squats recorded with a motion capture system (KneeKG?), and a quasi-static squat recorded with a low-dose upright biplanar radiographic imaging system (EOS®). The eight multi-joint models were evaluated by comparing dynamic knee kinematics optimized with GO to quasi-static knee kinematics measured by EOS®. The mean RMSE values ranged from 1.5-11.4° for rotations and 1.4-6.2 mm for translations. Some models were able to compensate for STA along one axe of movement but not along another. None of these models seemed optimal for STA compensation along all axes of movement. Future studies should investigate new models based on subject-specific joint geometry.