Cadaveric knees replaced with the Geomedic, ICLH, Marmor and Total Condylar prostheses were tested in axial compression, in rotation and in hyperextension in order to observe the strength of fixation of the tibial components. In axial compression the strengths at failure varied widely, both with any one prosthesis and between prostheses. This is attributed largely to the strength of the cancellous bone of the tibia, which was measured in each case and also varied widely. Three natural knees failed at loads of 7300, 7600 and 8300 newtons respectively, whereas the strengths of replaced knees ranged from 3000 to 15750 newtons. At least one example of each design failed at less than 7300 newtons, suggesting little or no reserve of strength. The strength of fixation was greater when the tibial prosthesis was large enough to rest on the whole cross-section of the tibia. In rotation the three prostheses embodying rollers in troughs were stiffer than the Marmor which had a nearly flat tibial-bearing surface. The presence or absence of the cruciate ligaments had a negligible effect on torsional stiffness. In hyperextension, knees replaced with the ICLH, Marmor and Total Condylar prostheses failed by rupture of the posterior capsule at moments of about 60 newton-metres, compared with about 100 for natural knees. With the Marmor prosthesis the anterior cruciate ligament was avulsed at about 20 newton-metres compared with about 75 in natural knees, suggesting that in this respect the retention of the cruciate ligaments contributes little. None of the four knees tested after inserting a Geomedic prosthesis showed strengths as high as those replaced with the other three designs.