A prospective, randomized trial of three antibiotic combinations (carbenicillin plus cephalothin, carbenicillin plus gentamicin, and cephalothin plus gentamicin) was designed for the treatment of suspected infection in 625 trials with granulocytopenic patients with cancer. Infection was documented in 396 episodes (63%), with gram-negative bacteremia in 86 (13%). Sixty-one percent of the isolated pathogens were gram-negative bacilli (principally Escherichia coli, Klebsiella species, and Pseudomonas aemginosa), and 16% were Staphylococcus aureus. The three antibiotic regimens employed had generally comparable degrees of efficacy. Overall results were excellent, and >70% of infected patients responded, although those with bacteremia, especially gram-negative bacteremia, responded less frequently than those without bacteremia. Provided that the pathogen was susceptible to the antibiotic (s) employed, the response of patients with gram-negative bacteremia depended primarily on whether their granulocyte count rose during therapy; 88% of the patients whose peripheral granulocyte count rose by >100 / μl improved, whereas only 22% of those whose granulocyte count did not rise improved. Although having a poor prognosis, this subgroup of patients with gram-negative bacteremia and persistent granulocytopenia represented only 7% of all infected granulocytopenic patients. The combination of cephalothin and gentamicin was found to be substantially more nephrotoxic (severe renal dysfunction in 12%) than the regimens of carbenicillin plus cephalothin (4%) and carbenicillin plus gentamicin (2%). Older patients whose creatinine levels were initially high normal had a 26% incidence of renal dysfunction related to treatment with cephalothin plus gentamicin. A combination of carbenicillin (or ticarcillin) and an aminoglycoside would currently appear to have the best ratio of efficacy to toxicity for initial empiric therapy of infection in granulocytopenic patients with cancer.