The influence of 27 variables on operative mortality and late complications (defined as residual or recurrent endocarditis or late bland periprosthetic leak) was determined using discriminant analysis for 108 patients undergoing valve replacement for native valve endocarditis at Stanford University Medical Center from March, 1964, to January, 1983. Congestive heart failure was the indication for valve replacement in 86% of patients. Aortic valve replacement was required in 68% and mitral valve replacement, in 26%. Patients were arbitrarily defined as having active (58%) or healed (42%) endocarditis. Follow-up included 515 patient-years and extended to a maximum of 19 years. Operative mortality was 15 +/- 4%, and 17 patients had late complications (linearized rate, 3.3% per patient-year). Seven variables were significantly related to operative mortality in the univariate analysis, but only organism (Staphylococcus aureus versus all others, p = 0.0302) was a significant independent predictor of operative mortality. For late complications, only 2 of 7 significant univariate covariates proved to be significant independent determinants: organisms on valve culture or gram stain and the presence of annular abscess. Patients with S. aureus endocarditis not showing prompt response to antibiotic treatment must be considered for early operation. Similarly, timely operative intervention for patients with annular abscess will be essential in decreasing late valve infections and perivalvular leaks.