Unicompartmental knee arthroplasty was performed in 207 knees of 179 patients using either a polycentric (188) or a geometric (nineteen) hemicomponent. After an average follow-up of 2.6 years, the results in 184 (89 per cent) of the knees were satisfactory. Pain and the need for ambulatory aids were reduced, and the distance the patients could walk was increased. Twenty-three (11 per cent) of the 207 procedures were rated as failures. The major cause of failure was loosening of the components (tibial in twelve and femoral in one); there also were unexplained pain in five knees, problems with the opposite unreplaced compartment in three, technical error in one, and pain in the patellofemoral joint in one. Nine of 155 intraoperative specimens for bacterial culture obtained during unicompartmental arthroplasty were positive, and two specimens that were positive on culture were obtained during revision of twenty failed unicompartmental arthroplasties. No gross or histological evidence of infection was demonstrated at operation. Based on this study, we concluded that this procedure can provide satisfactory relief of pain, adequate knee motion, and increased levels of independence and activity for patients with unicompartmental disease who are not suitable candidates for proximal tibial osteotomy or total knee replacement.