In a prospective, randomized multicenter trial of the Paul Ehrlich society different concepts for sequential empirical antimicrobial strategy for the treatment of patients with neutropenia less than 1.0/nl and fever greater than 38.5 degrees C and/or documented infection were studied. In phase I, patients with unexplained fever (FUO) were randomized for the combination of acylaminopenicillin plus aminoglycoside or third-generation cephalosporin plus aminoglycoside or double beta lactam therapy. Non-responders received additional vancomycin or all three substances of phase I in phase II. In phase III, all patients with persistent fever were then treated with amphotericin B plus 5-flucytosine and rifampin and randomized for the continuation of the double beta lactam regimen or additional imipenem/cilastatin. 667 (52.9%) of 1260 evaluable patients had FUO during the whole study period, of which 62.5% could be cured in phase I, 43.2% of non-responders in phase II and 55% of persistently febrile patients in phase III. The overall rate of complete response was 79.5%. 2.8% were non-responders, 11.7% were not evaluable for response and 40 patients (6%) died during the study, 24 (60%) of whom due to the underlying disease or the toxicity of antileukemic therapy. A significant difference between the treatment groups could not be detected in either of the three study phases.