Biomechanical models of the knee joint allow the development of accurate procedures as well as novel devices to restore the joint natural motion. They are also used within musculoskeletal models to perform clinical gait analysis on patients. Among relevant knee models in the literature, the anatomy-based spatial parallel mechanisms represent the joint motion using rigid links for the ligaments’ isometric fibres and point contacts for the articular surfaces. To customize analyses, therapies and devices, there is the need to define subject-specific models, but relevant procedures and their accuracy are still questioned. A procedure is here proposed and validated to define a customized knee model based on a spatial parallel mechanism. Computed tomography, magnetic resonance and 3D-video-fluoroscopy were performed on a healthy volunteer to define the personalized model geometry. The model was then validated by comparing the measured and the replicated joint motion. The model showed mean absolute difference and standard deviations in translations and rotations, respectively of $0.98\pm 0.40$ mm and $0.68\pm 0.29°$ for the tibia–femur motion, and of $0.77\pm 0.15$ mm and $2.09\pm 0.69°$ for the patella–femur motion. These results show that accurate personalized spatial models of knee kinematics can be obtained from in-vivo imaging.