From 1974 to 1986, 1430 cemented primary total knee arthroplasties were available for analysis. These included 224 total condylar prostheses with a polyethylene tibial component, 289 posterior stabilized prostheses with a polyethylene tibial component, and 917 posterior stabilized prostheses with a metal-backed tibial component. There were 12 failures in the total condylar series including three infections (one early and two late), five loose tibial components, two loose femoral and tibial components, and two unstable arthroplasties. The posterior stabilized prosthesis with the polyethylene tibia demonstrated six failures including two loose tibial components, two loose femoral components, and one supracondylar femur fracture. The posterior stabilized prosthesis with the metal-backed tibial component was associated with seven failures including six infections (three early and three late) and one femoral loosening. No metal-backed tibial components have been revised for loosening. The overall failure rate in this series was 1.7% for all prostheses. The incidence of tibial loosening was 0.53% with an average time to failure of 4.7 years. The incidence of femoral loosening was 0.35% with an average time to failure of 2.0 years. Tibial loosening was related to error in technique: postoperative varus tibiofemoral alignment, varus tibial component position, and excessive tibial resection. The overall infection rate was 0.63% for all total knee arthroplasties, and all were secondary to gram-positive organisms. Presently, the posterior stabilized prosthesis with a metal-backed tibia is the authors' implant of choice. Technical error and infection are the major causes of failure.