To determine which intensive care unit (ICU) infection rate may be best for interhospital and intrahospital comparisons and to assess the influence of invasive devices and type of ICU on infection rates, we analyzed data from the National Nosocomial Infections Surveillance System. From October 1986 to December 1990, 79 hospitals reported 2,334 hospital-months of data from 196 hospital units. The median overall infection rate was 9.2 infections per 100 patients. However, this infection rate had a strong positive correlation with average length of ICU stay (r = 0.60, p less than 0.0001). When patient-days was used in the denominator, the median overall nosocomial infection rate was 23.7 infections per 1,000 patient-days. Although there was a marked reduction in the correlation with average length of stay, this rate had a strong positive correlation with device utilization (r = 0.59, p less than 0.0001). To attempt to control for average length of stay and device utilization, we examined device-associated nosocomial infection rates. Central line-associated bloodstream infection rates, catheter-associated urinary tract infection rates, and ventilator-associated pneumonia rates varied by ICU type. The distributions of device-associated infection rates were different between some ICU types and were not different between others (coronary and medical ICUs or medical-surgical and surgical ICUs). Comparison of device-associated infection rates and overall device utilization identified hospital units with outlier infection rates or device utilization. These data show that: (1) choice of denominator is critical when calculating ICU infection rates; (2) device-associated infection rates vary by ICU type; and (3) intrahospital and interhospital comparison of ICU infection rates may best be made by comparing ICU-type specific, device-associated infection rates.