This study investigated the errors of obtaining visually selected anatomic landmarks for use in the registration process in a passive optical non-image-based computer-assisted total knee arthroplasty system in 5 fresh frozen cadavers. The projected maximum errors in the femoral mechanical axis (due to registration errors of the center of the distal femur) were 0.7 degrees in the coronal and 1.4 degrees in the sagittal plane. The projected maximum errors in the tibial mechanical axis arising from registration errors of the center of the proximal tibia were 1.3 degrees in the coronal and 2 degrees in the sagittal plane. The projected maximum errors in the transepicondylar axis were 9.1 degrees (registration errors of the medial femoral epicondyle) and 7.2 degrees (registration errors of the lateral femoral epicondyle). It should be noted that the results may be partly related to the use of the particular system in this experiment.