Twenty-nine knees with unicondylar, sixty-four with duocondylar, fifty with Guepar, and fifty with geometric prostheses were studied. The follow-up ranged from two to three and one-half years. The unicondylar prosthesis was used in the mildest cases and gave the least complications, but the quality of results was not superior to that achieved with the other prostheses. The duocondylar model was best suited for knees with rheumatoid arthritis and mild deformity. The geometric prosthesis was the best condylar prosthesis for osteoarthritis with moderate to severe deformity, but gave the worst results in knees with rheumatoid arthritis. The Guepar prosthesis was used in the worst knees and gave the best results, but it had the highest infection rate and was the most difficult to salvage. A radiolucency was observed in about 60 per cent of the condylar replacements around the tibial component and in 45 per cent of the Geupar replacements around the femoral component. The significance of this cannot yet be determined but it suggest that the fixation may not be ideal. In all types, residual pain was most frequently attributed to the patellar compartment. Patellectomy was not a solution.