To describe the epidemiology of nosocomial infections in medical intensive care units (ICUs) in the United States. Analysis of ICU surveillance data collected through the National Nosocomial Infections Surveillance (NNIS) System between 1992 and 1997. Medical ICUs in the United States. A total of 181,993 patients. Nosocomial infections were analyzed by infection site and pathogen distribution. Urinary tract infections were most frequent (31%), followed by pneumonia (27%) and primary bloodstream infections (19%). Eighty-seven percent of primary bloodstream infections were associated with central lines, 86% of nosocomial pneumonia was associated with mechanical ventilation, and 95% of urinary tract infections were associated with urinary catheters. Coagulase-negative staphylococci (36%) were the most common bloodstream infection isolates, followed by enterococci (16%) and Staphylococcus aureus (13%). Twelve percent of bloodstream isolates were fungi. The most frequent isolates from pneumonia were Gram-negative aerobic organisms (64%). Pseudomonas aeruginosa (21%) was the most frequently isolated of these. S. aureus (20%) was isolated with similar frequency. Candida albicans was the most common single pathogen isolated from urine and made up just over half of the fungal isolates. Fungal urinary infections were associated with asymptomatic funguria rather than symptomatic urinary tract infections (p < .0001). Certain pathogens were associated with device use: coagulase-negative staphylococci with central lines, P. aeruginosa and Acinetobacter species with ventilators, and fungal infections with urinary catheters. Patient nosocomial infection rates for the major sites correlated strongly with device use. Device exposure was controlled for by calculating device-associated infection rates for bloodstream infections, pneumonia, and urinary tract infections by dividing the number of device-associated infections by the number of days of device use. There was no association between these device-associated infection rates and number of hospital beds, number of ICU beds, or length of stay. There is a considerable variation within the distribution of each of these infection rates. The distribution of sites of infection in medical ICUs differed from that previously reported in NNIS ICU surveillance studies, largely as a result of anticipated low rates of surgical site infections. Primary bloodstream infections, pneumonia, and urinary tract infections associated with invasive devices made up the great majority of nosocomial infections. Coagulase-negative staphylococci were more frequently associated with primary bloodstream infections than reported from NNIS ICUs of all types in the 1980s, and enterococci were a more frequent isolate from bloodstream infections than S. aureus. Fungal urinary tract infections, often asymptomatic and associated with catheter use, were considerably more frequent than previously reported. Invasive device-associated infections were associated with specific pathogens. Although device-associated site-specific infection rates are currently our most useful rates for performing comparisons between ICUs, the considerable variation in these rates between ICUs indicates the need for further risk adjustment.