More than 90% of all intravascular device-related septicaemias are due to central venous or arterial catheters. To assess the efficacy of cutaneous antisepsis to prevent catheter-associated infection, we prospectively studied three antiseptics for disinfection of patients' central venous and arterial catheter insertion sites in a surgical intensive care unit. 668 catheters were randomised to 10% povidone-iodine, 70% alcohol, or 2% aqueous chlorhexidine disinfection of the site before insertion and for site care every other day thereafter. Chlorhexidine was associated with the lowest incidence of local catheter-related infection (2.3 per 100 catheters vs 7.1 and 9.3 for alcohol and povidone-iodine, respectively, p = 0.02) and catheter-related bacteraemia (0.5 vs 2.3 and 2.6). Of the 14 infusion-related bacteraemias (4 due to contaminated infusate or catheter hub, 10 due to infected catheters), 1 was in the chlorhexidine group and 13 were in the other two groups (odds ratio 0.16, p = 0.04). We conclude that use of 2% chlorhexidine, rather than 10% povidone-iodine or 70% alcohol, for cutaneous disinfection before insertion of an intravascular device and for post-insertion site care can substantially reduce the incidence of device-related infection.