We studied 300 episodes of infective endocarditis in 287 patients to evaluate clinical and laboratory determinants of prognosis by estimating the probability of in-hospital death. The patients' ages ranged from 2 months to 78 (mean 30.76 +/- 16.06) years; 185 (62%) episodes occurred in male and 115 (38%) in female patients. A total of 386 complications occurred in 223 (74%) episodes of endocarditis. The infecting microorganisms were streptococci in 147 episodes, Staphylococcus aureus in 59, Staphylococcus epidermidis in 14, gram-negative bacteria in 16, other gram-positive bacteria in 8, fungi in 4. The causative microorganism was not identified in 52 episodes (negative blood cultures). The underlying cardiac disease was valvular in 119 episodes, congenital in 37, prosthetic heart valve in 69, and others in 6. No previous heart disease was identified in 69 episodes. Surgical treatment was carried out in 102 (34%) patients. Overall, 78 (26%) patients died. The probability of death was estimated with a logistic regression model (stepwise procedure). The model with best prediction included the cardiac status previous to the endocarditis, the causative microorganism, the occurrence of complications, and the blood leukocyte count. The most important variable in predicting in-hospital death was the occurrence of complications, followed by cardiac status (prosthetic valve endocarditis), the infecting microorganism, and leukocyte count. The model underestimated the severity of the disease in patients with acute endocarditis and overestimated in patients with prosthetic valve endocarditis submitted to surgical treatment.