Primary nosocomial bloodstream infection (BSI) is a common occurrence in the intensive care unit (ICU) and is associated with a crude mortality of 31.5 to 82.4%. However, an accurate estimate of the attributable mortality has been limited because of confounding by severity of illness. We undertook this study to assess the attributable mortality and costs associated with an episode of BSI. Infected patients were defined as those who had an episode of BSI during the study period. Uninfected control subjects were matched to the infected patients based upon a number of factors, including predicted mortality on the day prior to infection. The main outcome measures were crude ICU mortality, length of stay, and costs. We found no difference in the crude mortality for the infected and the uninfected patients (35.3 and 30.9%, respectively, p = 0.51). However, among survivors, the patients with nosocomial bloodstream infections did have excess length of stay (mean, 10 d; median, 5 d; p = 0.007) and increased direct costs (mean difference, $34,508; p = 0.008). After matching for severity of illness, we could not detect an association between primary nosocomial bloodstream infections and increased ICU mortality. We did find that primary nosocomial bloodstream infections increased ICU length of stay and costs.