To determine risk factors for Serratia marcescens infection or colonization, and to identify the source of the pathogen and factors facilitating its persistence in a neonatal intensive-care unit (NICU) during an outbreak. Retrospective case-control study; review of NICU infection control policies, soap use, and handwashing practices among healthcare workers (HCWs); and selected environmental cultures. A university-affiliated tertiary-care hospital NICU. All NICU infants with at least one positive culture for S marcescens during August 1994 to October 1995. Infants who did not develop S marcescens infection or colonization were selected randomly as controls. Thirty-two patients met the case definition. On multivariate analysis, independent risk factors for S marcescens infection or colonization were having very low birth weight (< 1,500 g), a patent ductus arteriosus, a mother with chorioamnionitis, or exposure to a single HCW. During January to July 1995, NICU HCWs carried their own bottles of 1% chlorxylenol soap, which often were left standing inverted in the NICU sink and work areas. Cultures of 16 (31%) of 52 samples of soap and 1 (8%) of 13 sinks yielded S marcescens. The 16 samples of soap all came from opened 4-oz bottles carried by HCWs. DNA banding patterns of case infant, HCW soap bottle, and sink isolates were identical. Extrinsically contaminated soap contributed to an outbreak of S marcescens infection. Very-low-birth-weight infants with multiple invasive procedures and exposures to certain HCWs were at greatest risk of S marcescens infection or colonization.