Objective To build a 2D/3D registration system based on the compute unified device architecture(CUDA) frame with single X-ray image and CT data of knee joints and apply it in the research of knee motion and stability of implanted prosthesis. Method The digital radiography(DR) equipment used in the study was calibrated by the Zhang zhengyou Calibration Method, and then digitally rendered radiographs(DRR) images were generated in the CUDA frame with lighttracing algorithm, and the best 2D/3D registration parameters were calculated with a similarity operator of cross correlation; finally, the results were evaluated by using the method of 3D/3D registration with data obtained from a 3D laser scanner. Results With knee specimen X-ray images and CT data, in 6 degrees of freedom, the average errors of transform were below 1 mm, and those of rotation were below 1°. Conclusions The 2D/3D registration system can meet the precision requirement of motion detection and be used to study the knee motion and prosthesis location.