We conducted a prospective, randomized, multicentric study in community hospitals. Patients with clinical sepsis, rectal temperature > or = 38 degrees C and pulse rate > or = 100 bpm were randomized to receive ceftazidime (group CAZ) or a combination of antibiotics freely chosen by the clinician following his 'best guess' (group COMB). On specified grounds, the clinician could also treat patients in an open group with a free combination of antibiotics (group OPEN). The severity of disease at study admission was assessed by a clinical estimation and an Apache II score. There were 128 patients included: 56 randomized in group CAZ, 50 in group COMB, and 22 in the OPEN group. Ninety-one patients were evaluable: 41 in group CAZ, 30 in group COMB, 20 in OPEN group. At the end of the period of empirical treatment (48-72 h), the clinical success rates (improvement of status) were 93, 93 and 75% (p for group OPEN vs. groups CAZ or COMB: 0.10). The bacteriological success rates (sterile blood cultures) were 91, 88 and 80% (p not significant). mean Apache II score was 16.7 and the score correlated significantly with outcome, as did clinical evaluation. In conclusion, ceftazidime alone was a safe antibiotic therapy in this study and we could not demonstrate a superiority of a combined antibiotic therapy chosen by the clinician following his 'best guess' over ceftazidime.