We examined the in-vitro activity of 12 antibiotics against Gram-negative bacillary isolates from 141 distinct episodes of nosocomial bloodstream infection occurring from July 1984 through November 1986. At least ten strains of each of the seven most frequently encountered species were tested. Relative potency was carefully assessed by extending the concentrations from 0.004 to 64 mg/l in microdilution tests performed in duplicate. We estimated MIC50 and MIC90 and, importantly, calculated 95% confidence intervals (CI95) for MIC90. Against all isolates, ciprofloxacin, the most potent antibiotic, had an MIC50 of 0.03 and an MIC90 of 0.13 (CI95 0.11 to 0.16 mg/l). Norfloxacin and enoxacin had MIC90 of 0.50 and 0.71 mg/l, respectively. Imipenem, ceftazidime, aztreonam, and cefoperazone had MIC90 of 1.0, 2.0, 5.3, and 5.7 mg/l, respectively. Both cefotaxime and ceftriaxone had MIC90 of 16 mg/l (CI95 13-19.7 mg/l). For tobramycin, gentamicin and amikacin, the MIC90 values were 1.4, 5.7, and 8 mg/l, respectively. Against Pseudomonas aeruginosa (n = 26), Serratia marcescens (n = 19), and Klebsiella pneumoniae (n = 26), the CI95s about the MIC90 for ciprofloxacin were 0.31-0.81, 0.07-0.23, and 0.04-0.10 mg/l, respectively. For optimal comparison of antibiotics used to treat hospital-acquired bacteraemias, only clinically significant nosocomial bloodstream isolates should be studied with regard to their antibiotic susceptibilities; the isolates should be unique (only one isolate per episode of bacteraemia occurring over a defined period of time); an adequate number of isolates of a particular species should be studied; MICs should be determined over a wide range of concentrations; and both the MIC90 and the CI95 should be reported.