This study assessed the three-dimensional accuracy of Magnetic Resonance Imaging (MRI) for measuring articular surface topographies and cartilage thicknesses of human cadaveric knee joints, by comparison with a calibrated stereophotogrammetric (SPG) method. Three-dimensional spoiled gradient recalled acquisition with fat suppression was employed with a linear extremity coil in a 1.5 T superconducting magnet and an imaging voxel size of 0.47 x 0.47 x 1.0 mm. For six cadaveric knees, the average accuracies of cartilage and subchondral bone surface measurements were found to be 0.22 mm and 0.14 mm respectively; the thickness measurements demonstrated an average accuracy of 0.31 mm. It was found that while most of the error may be attributed to random measurement error, the accuracy was in fact diminished somewhat by systematic errors. For each bone of the knee, accuracies were most favorable in the patella, followed by the femur then the tibia. This study employed a manual and a semi-automated segmentation method for extracting topographic measurements from MRI; it was found that the much more efficient semi-automated method provided equally good and sometimes better accuracies than manual segmentation. Finally, it was demonstrated that methodologies employed previously in the analysis of SPG data could be applied to measure cartilage topography, thickness, contact areas and surface curvatures from in vivo MRI measurements of the knees of four volunteers.