To determine the excess mortality attributable to hospital-acquired bloodstream infections, we applied the acute physiology and chronic health evaluation (APACHE) II classification to 34 critically ill patients with this complication. The study included primary bloodstream infections, defined by a positive blood culture at least three days after hospitalization, in the absence of any other apparent source of infection. The most frequent blood isolates included Staphylococcus aureus (39 percent), Gram-negative rods (24 percent), and Candida albicans (15 percent); the spectrum of blood isolates suggested that most infections were related to intravascular catheters. In a control group of intensive care unit patients (n = 384), the death rate predicted by APACHE II was similar to the observed death rate (35.3 vs 37.8 percent). In a subgroup of control patients (n = 34), chosen for APACHE II scores that matched the patients with bloodstream infections, predicted and observed death rates were also similar (53.1 vs 52.9 percent). For patients with bloodstream infections, however, observed mortality (82.4 percent) significantly exceeded the predicted value (54.1 percent, p = 0.025). We conclude that critically ill patients who develop nosocomial bloodstream infections are at greater risk of death than patients with comparable severity of illness without this complication. The difference between the observed and predicted death rates, 28 percent, represents the excess mortality associated with bloodstream infection in critically ill patients.