HYPOTHESIS: A set of clinical variables available at the bedside can be used to predict outcome in critically ill patients with bloodstream infection (BSI). DESIGN: A 3-year retrospective cohort study. SETTING: A surgical intensive care unit in Switzerland. PATIENTS: All patients with BSI were potentially eligible. MAIN OUTCOME MEASURES: Clinical variables, organ dysfunctions, and outcome. RESULTS: Among 4530 admissions to the surgical intensive care unit, 224 clinically significant episodes of BSI were recorded (incidence, 4.9%), with a 28-day fatality of 36%. A total of 110 patients had primary bacteremia, of which 39 (35%) were catheter related. Although gram-positive organisms were the most frequently isolated pathogens (58% [159/275]), they were associated with lower case-fatality (30%) than BSI due to gram-negative bacteria (44%). Organ dysfunctions associated with the highest risk of death were neurologic dysfunction (hazard ratio [HR], 6.9; 95% confidence interval [CI], 3.3-14.5), hepatic dysfunction (HR, 3.9; 95% CI, 2.1-7.4), and disseminated intravascular coagulation (HR, 3.0; 95% CI, 1.5-6.1). By multivariate analysis, 2 independent predictors of mortality were the APACHE II (Acute Physiology and Chronic Health Evaluation II) score at onset of BSI (HR per 1-point increase, 1.08; 95% CI, 1.04-1.12) and the number of evolving organ dysfunctions (HR, 1.4; 95% CI, 1.2-1.7). Appropriate antimicrobial therapy was associated with improved outcome (HR, 0.4; 95% CI, 0.2-0.6). CONCLUSIONS: Bloodstream infection in critically ill patients is a common and frequently fatal condition. Its outcome can be predicted by the severity of illness at onset of BSI and the number of organ dysfunctions evolving thereafter. Appropriate antimicrobial therapy is an important determinant for survival.