To achieve quantitative assessment of 3D dynamic motion of artificial knee implants under clinical conditions, we developed a 3D kinematic analysis system using X-ray fluoroscopic imaging. The 3D pose-estimation technique for knee implants was built on a 2D/3D registration algorithm, which determines the spatial pose for each femoral and tibial component from the knee implant contours and computer-assisted design (CAD) models of the implant. In order to validate the accuracy of the 3D pose estimation and the system, computer simulation and in vitro tests were performed using images of knee implants taken in 10 different poses with respect to X-ray focus. Computer simulation tests showed that the root mean square errors (RMSE) for all variables were less than 1.0 mm 1.0 degrees. In vitro tests showed that the RMSE for translation perpendicular to the X-ray image plane was about 1.5 mm, while the accuracy of the remaining two translational and three rotational variables was found to be sufficient for analyzing knee kinematics. Computation time in 3D pose estimation was then obtained in less than 30 seconds for each frame. In clinical application, dynamic movement in deep knee bending was quantitatively analyzed, and the feasibility and effectiveness of the system was demonstrated.