To determine whether combination antibiotic therapy including a short course of an aminoglycoside was as effective and less toxic than a conventional long course of the combination for the empirical therapy of gram-negative bacteremia in patients with cancer and granulocytopenia, we conducted a randomized multicenter trial comparing ceftazidime plus a short course (three days) of amikacin, ceftazidime plus a long course (nine days) of amikacin, and azlocillin plus a long course (nine days) of amikacin. Single-organism gram-negative bacteremia occurred in 129 of 872 evaluable patients. Without a change in antibiotics, the response rates were 81 percent with ceftazidime and long-course amikacin, 48 percent with ceftazidime and short-course amikacin (P = 0.002), and 40 percent with azlocillin and long-course amikacin (P less than 0.001). Among patients with fewer than 100 granulocytes per cubic millimeter throughout therapy, the response rates were 6 percent with ceftazidime and short-course amikacin and 50 percent with ceftazidime and long-course amikacin (P = 0.03). Linear logistic-regression analysis showed that therapy with ceftazidime and long-course amikacin was the most favorable prognostic factor of the response to infection, whereas the presence of leukemia or shock was the least favorable. We conclude that ceftazidime should be given in combination with a conventional full course of an aminoglycoside (amikacin) when used for the empirical treatment of gram-negative bacteremia in cancer patients with granulocytopenia.