OBJECTIVES: To determine increased hospital stay and direct costs attributable to hospital-acquired, laboratory-confirmed bloodstream infection (BSI), and to evaluate the matching variable length of stay (LOS). DESIGN: Retrospective (historical) cohort study with 1:2 matching in intensive care units and surgical wards. SETTING: A 2,000-bed university hospital in Rome, Italy. PATIENTS: All patients admitted between January 1994 and June 1995 who had hospital-acquired, laboratory-confirmed BSI were considered cases; all others were eligible as controls. METHODS: Two controls (A and B) were selected per case in a stepwise fashion. Controls in group A were selected according to the following six criteria: ward, gender, age, diagnosis, central venous catheter, and LOS equal to the interval from admission to infection in a matched case +/- 20% (LOS +/- 20%). Controls in group B were selected according to the first five criteria, but excluded LOS +/- 20%. RESULTS: One hundred five of 108 patients were each matched with two controls. The matching appropriateness score was greater than 90%. With the use of controls in groups A and B, the case-fatality rates attributable to hospital-acquired, laboratory-confirmed BSI were 35.2% and 40.9%, respectively; the estimated risk ratios for death were 2.60 and 3.52 (P = .0001), respectively. The increased hospital stay per case attributable to hospital-acquired, laboratory-confirmed BSI was 19.1 (mean) and 13.0 (median) days for matched pairs in control group A and 19.9 (mean) and 15.0 (median) days for matched pairs in control group B. With controls in group A, the cost of increased hospital stay per patient attributable to hospital-acquired, laboratory-confirmed BSI was Euro 15,413. The additional cost per patient due to treatment was Euro 943, making the overall direct cost Euro 16,356 per case. CONCLUSIONS: This study should make it possible to estimate the cost of hospital-acquired, laboratory-confirmed BSI in most hospitals after adjusting for incidence rate. It also confirmed the use of LOS +/- 20% as a matching variable to limit overestimation of increased hospital stay. To our knowledge, this is among the first such studies in Europe.