The experience of 810 mitral valve replacements performed during a 15-year period (1969 to June 1983) is reported. All but one patient received a Björk-Shiley disc valve. Five different Björk-Shiley models were used (Delrin, n = 51; standard pyrolytic carbon, n = 307; convexo-concave with an opening angle of 60 degrees, n = 229; convexo-concave with an opening angle of 70 degrees, n = 123; and the Monostrut valve, n = 99). The follow-up was 99.8% complete and covered 4614 patient-years (mean 6.0 years). An autopsy was performed in 77% of all fatalities. Early mortality was 5.7% and decreased during the study period. The 5-, 10-, and 15-year actuarial survival rates (early mortality excluded) were 78%, 62%, and 51%, respectively, and the complication-free survival rates at the same time intervals were 67%, 51%, and 36%. Seven percent of early deaths and 25% of late deaths were valve related, and most of these were due to valve thrombosis. The incidence of valve thrombosis was significantly (p less than 0.0005) reduced after the introduction of the convexo-concave valves, and this reduction was achieved without any increase in bleeding complications. The only patient-related factor found to be related to postoperative thromboembolism was a history of preoperative embolism (p less than 0.05). Most events occurring after implant (death, embolism, valve thrombosis, reoperation, and valve failure) had a peak incidence during the first postoperative year, whereas anticoagulant-related bleedings occurred at a constant rate of 1.2/100 patient-years. The linearized incidence of embolism was 1.6/100 patient-years (actuarial incidence at 5 and 10 years 6.9% and 13.0%, respectively). Seven strut fractures occurred (0.2/100 patient-years, actuarial incidence at 5 and 10 years 1.1%), as did five instances of prosthetic valve endocarditis (0.1/100 patient-years, actuarial incidence at 5 and 10 years, 0.7% and 1.3%). A new concept of valve failure is presented and includes a distinction between events related to the surgical procedure (treatment failure; 1.9/100 patient-years, actuarial incidence at 5 and 10 years 11.0% and 13.5%) and events possibly related to the prosthesis (prosthetic failure; 1.1/100 patient-years, actuarial incidence at 5 and 10 years 7.0% and 8.0%).