To improve identification of children excreting rotavirus a method for the amplification of rotavirus RNA by the polymerase chain reaction (PCR) was developed. The assay was compared with a solid-phase enzyme immunoassay in the detection of rotavirus shedding by infants in hospital during the winter peak of rotavirus infections. Forty children were studied in an intermediate care unit after transfer from intensive care units. Only two were admitted primarily because of diarrhoea; the other thirty-eight were admitted for management of various other disorders. Rotavirus shedding was detected by enzyme immunoassay in twenty of the infants, and nine of these (aged 1 week to 8 months) remained in hospital for more than 5 days after the initial detection of rotavirus and could be studied long term. Of 103 faecal samples from the nine infants, 60 (58%) contained rotavirus RNA detected by reverse-transcriptase (RT)/PCR, whereas only 37 (36%) were positive for rotavirus antigen by the immunoassay (chi 2 = 10.3, p less than 0.002). The geometric mean time of rotavirus shedding was 9.5 (range 1-19) days as detected by RT/PCR and 5.7 (range 1-17) days by the immunoassay (p less than 0.018). In five of the nine children, RT/PCR detected rotavirus shedding for 2-7 days longer than the immunoassay and in four children RT/PCR was positive 1 or more days before rotavirus antigen was detected. Further studies should attempt to find out whether infected infants are capable of spreading wild-type virus during periods when they are not shedding antigen as detectable by enzyme immunoassay.