If nosocomial infections are to be used as clinical indicators of quality, their definitions must be accurate. To assess validity and reliability of definitions of nosocomial infection, a study was conducted in two groups of U.S. hospitals. Group A consisted of a stratified, random sample of 715 hospitals and excluded those that are part of the National Nosocomial Infections Surveillance System. The 112 NNIS hospitals were surveyed separately in group B. Both groups used the same instrument, consisting of 36 case studies simulating patients' charts. Content and construct validity were formally tested and demonstrated. Six case studies were presented for each of the four major NI sites and for community-acquired or no infection. The pooled hospital response was 48% (396/827). The pooled number of individual responders whose data were used in the analysis was 469. Their overall mean score was 84%, and the score for correctly identifying any NI was 83%. Both groups were best at identifying urinary nosocomial infections (Group A = 92%, Group B = 93%) and poorest for no infection (Group A = 62%, Group B = 75%). Group A responders had significantly higher scores if they were certified, had a baccalaureate or higher degree, had taken a formal infection control course, had worked in infection control for greater than or equal to 2 years, or had worked full time in infection control in a greater than or equal to 200-bed hospital that was affiliated with a medical school (all p less than 0.05).(ABSTRACT TRUNCATED AT 250 WORDS)