Calculating knee kinematics from trajectories of markers glued on the skin of the lower limbs, or mounted on specific tools attached to the thigh and shank, is a frequently used technique in gait analysis. However, this technique suffers from large inaccuracies caused by soft tissue artefacts (STA) (Peters et al. 2010). Various methods were developed to compensate for STA, but none seems satisfactory (Leardini et al. 2005). Among these methods, global optimisation (GO) provides the optimised position of the lower limbs by minimising the differences between measured and model-predicted markers coordinates under kinematic constraints (Lu et al. 1999). Many multi-joint models were tested with GO (Duprey et al. 2010), but few of them were directly validated on subjects (Stagni et al. 2009 ; Andersen et al. 2010 ; Li et al. 2012 ). Moreover, GO has never been validated in osteoarthritis (OA) subjects. The goal of this study was, therefore, to evaluate the performance of eight multi-joint models used in GO to compensate for STA on healthy and OA subjects performing squats.