A computerized method to automatically and spatially align joint axes of in vivo knee scans was established and compared to a fixed reference system implanted in a cadaver model. These computational methods to generate geometric models from static MRI images with an automatic coordinate system fitting proved consistent and accurate to reproduce joint motion in multiple scan positions. Two MRI platforms, upright and closed, were used to scan a phantom cadaver knee to create a three-dimensional, geometric model. The knee was subsequently scanned in several positions of knee bending in a custom made fixture. Reference markers fixed to the bone were tracked by an external infrared camera system as well as by direct segmentation from scanned images. Anatomical coordinate systems were automatically fitted to the segmented bone model and the transformations of joint position were compared to the reference marker coordinate systems. The tracked translation and rotation measurements of the automatic coordinate system were found to be below root mean square errors of 0.8 mm and 0.7°. In conclusion, the precision of the translation and rotational tracking is found to be sensitive to the scanning modality, albeit in upright or closed MRI, but still within comparative measures to previously performed studies. The potential to use segmented bone models for patient joint analysis could vastly improve clinical evaluation of disorders of the knee with continual application in future three-dimensional computations.