The aim o f this m ulticentre randomised trial was to determine whether it was possible to predict grampositive bacteraem ia, and whether the em pirical use o f vancomycin would lead to reduced morbidity and mortality. 35 o f 113 patients (31%; confidence interval. Cl 8.5), who presented with a skin or soft tissue infection and had received empirical vancomycin in addition to either ceftazidime or piperacillin-tobram ycin, had initial bacteraemia with a single gram-positive bacterium compared with 135 of the 784 (17%; C l 2.6), who presented with another infection and who had been given ceftazidime or piperacillin-tobram ycin without vancomycin (P<0.G01). Empirical vancomycin resulted in a higher rate of eradication (P = 0.033, relative risk 1.2), but not a better clinical outcome and was associated with more toxicity (P=0.042, relative risk 1.6). Irrespective of the initial treatment regimen, fever lasted an average of 8 days, the em pirical regimen was modified in more than 50% of cases and mortality attributed to gram-positive infection was less than 2%. Incorporating vancomycin in the initial em pirical antibiotic regimen for febrile neutropenic patients does not appear necessary, even for skin and soft tissue infections associated with gram -positive bacteraemia.