Selective decontamination of the digestive tract (SDD), by means of non-absorbable antibiotics, to prevent infection in intensive-care units (ICUs) remains controversial; there is evidence that the regimen reduces the incidence of secondary infection, but no convincing reduction in morbidity or mortality has been shown and the costs and effect on microbial resistance patterns need further study. In a double-blind, placebo-controlled trial, we have tried to find out whether SDD should be used routinely in all ICU patients at high risk of secondary infection. All patients admitted to the ICU who were thought likely to stay in the unit for at least 5 days and to need intubation for longer than 48 h were enrolled and randomly allocated to groups receiving placebo or SDD (amphotericin, colistin, and tobramycin applied to the oropharynx and enterally); all patients received intravenous cefotaxime for 72 h. Of 322 patients randomised, 83 were withdrawn (80 ICU stay or duration of intubation too short, 3 protocol violations). 239 medical, trauma, and surgical patients completed the trial period (114 SDD, 125 placebo). There were no differences between SDD and placebo groups in incidence of infection (30 [26%] vs 43 [34%] patients; p = 0.22), duration of ICU stay (mean 16.2 [14.3] vs 16.8 [12.3] days), hospital stay (29.9 [SD 25.0] vs 31.9 [22.2] days), or mortality (21 [18%] vs 21 [17%]). SDD substantially increased the costs of intensive care. Mechanisms other than bacterial colonisation of the gut may bring about substantial numbers of secondary infections in ICUs. Routine use of SDD in multidisciplinary ICUs cannot be recommended.