© 2017 Elsevier Ltd Background Continuation of empirical antimicrobial therapy (EAT) for febrile neutropenia in patients with haematological malignancies until neutrophil recovery could prolong the therapy unnecessarily. We aimed to establish whether EAT discontinuation driven by a clinical approach regardless of neutrophil recovery would optimise the duration of therapy. Methods We did an investigator-driven, superiority, open-label, randomised, controlled phase 4 clinical trial in six academic hospitals in Spain. Eligible patients were adults with haematological malignancies or haemopoietic stem-cell transplantation recipients, with high-risk febrile neutropenia without aetiological diagnosis. An independent, computer-generated randomisation sequence was used to randomly enrol patients (1:1) to the experimental or control group. Investigators were masked to assignment only before randomisation. EAT based on an antipseudomonal β-lactam drug as monotherapy (ceftazidime or cefepime, meropenem or imipenem, or piperacillin-tazobactam) or as combination therapy (with an aminoglycoside, fluoroquinolone, or glycopeptide) was started according to local protocols and following international guidelines and recommendations. For the experimental group, EAT was withdrawn after 72 h or more of apyrexia plus clinical recovery; for the control group, treatment was withdrawn when the neutrophil count was also 0·5 × 109 cells per L or higher. The primary efficacy endpoint was the number of EAT-free days. Primary analyses were done in the intention-to-treat population. Efficacy and safety analyses were done in the intention-to-treat population and the per-protocol population. This trial is registered with ClinicalTrials.gov, number NCT01581333. Findings Between April 10, 2012, and May 31, 2016, 157 episodes among 709 patients assessed for eligibility were included in analyses. 78 patients were randomly assigned to the experimental group and 79 to the control group. The mean number of EAT-free days was significantly higher in the experimental group than in the control group (16·1 [SD 6·3] vs 13·6 [7·2], absolute difference −2·4 [95% CI −4·6 to −0·3]; p=0·026). 636 adverse events were reported (341 in the experimental group vs 295 in the control group; p=0·057) and most (580 [91%]; 323 in the experimental group vs 257 in the control group) were considered mild or moderate (grade 1–2). The most common adverse events in the experimental versus the control group were mucositis (28 [36%] of 78 patients vs 20 [25%] of 79 patients), diarrhoea (23 [29%] of 78 vs 24 [30%] of 79), and nausea and vomiting (20 [26%] of 78 vs 22 [28%] of 79). 56 severe adverse events were reported, 18 in the experimental group and 38 in the control group. One patient died in the experimental group (from hepatic veno-occlusive disease after an allogeneic haemopoietic stem-cell transplantation) and three died in the control group (one from multiorgan failure, one from invasive pulmonary aspergillosis, and one from a post-chemotherapy intestinal perforation). Interpretation In high-risk patients with haematological malignancies and febrile neutropenia, EAT can be discontinued after 72 h of apyrexia and clinical recovery irrespective of their neutrophil count. This clinical approach reduces unnecessary exposure to antimicrobials and it is safe. Funding Instituto de Salud Carlos III, Spanish Ministry of Economy (PI11/02674).