In December 1999, an outbreak of diarrhea was reported in a general hospital neonatal medium care unit (NMCU) caused by a novel strain of rotavirus with genotype P[6], G9. An investigation was conducted to determine risk factors for illness among neonates. Rotavirus diagnosis was by latex agglutination and typing by reverse transcriptase polymerase chain reaction. A case-control study was performed using data collected from medical records on exposures in a 3-day period before illness (cases) or a random 3-day period (controls). Environmental swabs were tested for rotavirus. Antenatal blood samples from mothers and blood samples provided by hospital staff were analyzed for rotavirus antibodies. Fifty-six cases of rotaviral illness were confirmed by latex agglutination. Forty-seven of these were among 118 neonates exposed to the NMCU (attack rate, 40%). There was a 4-week period with no clinical cases in the course of the outbreak. Increased frequency (> or = 15 times in 3 days) of ungloved nasogastric feeding was a significant risk factor (adjusted odds ratio, 8.79), controlling for birth weight and gestational age. Environmental sampling showed persistence of the virus on ward surfaces despite cleaning. None of 24 NMCU staff members had high levels of antibodies against P[6], G9. Three (8%) of 38 mothers had high antibody levels; 2 had infants who became ill. The outbreak ended with a 7-day ward closure, disinfection, and introduction of gloved nasogastric feeding. Case-control studies can be successful in identiffying risk factors for nosocomial outbreaks of diarrhea. High levels of rotavirus antibodies in mothers may not protect infants. The environment may be the most important reservoir of rotavirus during outbreaks.