One thousand six hundred and sixty-eight consecutive patients who underwent isolated mitral valve replacement (MVR) from 1963 to 1984 were evaluated retrospectively. Thromboembolism occurred with a linearised rate of 2.5% +/- 0.2%/patient-year (PY) for Starr-Edwards disc prosthesis Model 6520, 2.4% +/- 0.3%/PY for Bjørk-Shiley plane prosthesis, 3.0% +/- 0.8%/PY for Bjørk-Shiley convexo-concave 60 degrees prosthesis, 3.0% +/- 0.8%/PY for St. Jude Medical prosthesis and 3.4% +/- 0.5%/PY for Carpentier-Edwards tissue valve without the differences reaching significance. In the SJM group, the incidence of thromboembolism was significantly higher (P less than 0.025) in smaller sizes (less than M29) probably due to a more turbulent flow. The linearised rate for major haemorrhage was 1.6% +/- 0.1%/PY. Twenty-three percent of the thromboembolic and 18% of the bleeding events were fatal. Sixty-eight percent of the emboli involved the central nervous system and bleeding apart from fatalities was predominantly non-cerebral (81%). Whereas thromboembolism was a time-related event with more than twice as high a risk in the first postoperative year (4.2% +/- 0.5% vs. 1.7% +/- 0.8%, P less than 0.01), bleeding occurred with a constant rate over time (0.9% +/- 0.4%). Adequacy of anticoagulation was an important risk factor for postoperative embolism with the prothrombin time (PT) exceeding the therapeutic range in 65% of all events. A preoperative history of embolism was the only additional patient-related risk factor for postoperative embolism (18.3% vs. 9.6%, P less than 0.001). In 30% of all haemorrhage, the PT was below 15%.(ABSTRACT TRUNCATED AT 250 WORDS)