When estimating knee kinematics from skin markers and stereophotogrammetry, multi-body optimization (MBO) has provided promising results for reducing soft tissue artefacts (STA), but can still be improved. The goal of this study was to assess the performance of MBO with subject-specific knee models at high knee flexion angles (up to 110°) against knee joint kinematics measured by magnetic resonance imaging. Eight subjects were recruited. MBO with subject-specific knee models was more effective in compensating STA compared to no kinematic and spherical constraints, in particular for joint displacements. Moreover, it seems to be more reliable over large ranges of knee flexion angle. The ranges of root mean square errors for knee rotations/displacements were 3.0°-9.2°/1.3-3.5 mm for subject-specific knee models, 6.8°-8.7°/6.0-12.4 mm without kinematic constraint and 7.1°-9.8°/4.9-12.5 mm for spherical constraints.