Surveillance activities for detection of nosocomial infections at the University of Virginia Hospital (Charlottesville, Va.) and at hospitals participating in the Virginia Statewide Infection Control Program have focused on device-related infections, which potentially are preventable. Surveillance data collected between January 1, 1975, and December 31, 1979, indicated that (1) a high proportion (33%-45%) of all nosocomial bloodstream infections occurred among patients in intensive care units (ICUs), who occupied only 8% of hospital beds; (2) patients in the burn unit, neonatal ICU, and surgical ICU had a 24-fold, lO-fold, and eightfold higher rate of nosocomial bloodstream infection, respectively, than did patients on general wards; (3) up to 73010 of surgical ICU patients had at least one intravascular device besides an iv catheter, and a high proportion (3%-17 %) of these patients subsequently developed a nosocomial bloodstream infection; (4) ICU patients constituted a large proportion (31%) of those patients infected with gram-negative rods whose infecting organism was resistant to aminoglycoside antibiotics; (5) five major outbreaks, which involved 110 patients (4% of infected patients), were identified at the University Hospital over a period of two years; all of the outbreaks involved ICU patients, and three outbreaks directly involved devices; and (6) the efficiency of surveillance for life-threatening infections was greatest in ICUs. With the recognition that ICU patients are at high risk of acquiring nosocomial infections, and with the knowledge that a significant proportion of infections are associated with the use of devices and are thus preventable, the highest priority for infection control resources should be assigned to surveillance of patients in ICUs.