Background: Early antibiotic discontinuation has been advocated in haematology patients with fever of unknown origin during chemotherapy-induced neutropenia, but its safety is unknown. We aimed to assess if short treatment with carbapenems is non-inferior to extended treatment. Methods: This non-inferiority, open-label, multicentre, randomised trial was done in six hospitals in the Netherlands. Adult patients (≥18 years) who were treated with intensive chemotherapy or haematopoietic stem-cell transplantation (HSCT) for a haematological malignancy, and had fever of unknown origin during high-risk neutropenia (<0·5 × 10 9/L expected for ≥7 days) were eligible. After onset of fever, patients received either 500 mg intravenous imipenem-cilastatin four times a day or 1000 mg intravenous meropenem three times a day. Between 48 h and 72 h of treatment, participants were randomly assigned (1:1) by a computer-generated sequence to receive a short-term (72 h [60–84]; short treatment group) or extended (≥9 days until being afebrile for 5 days or neutrophil recovery; extended treatment group) carbapenem regimen. The composite primary endpoint was treatment failure, defined as recurrent fever or a carbapenem-sensitive infection between day 4 and day 9 and septic shock or respiratory failure or death from day 4 until neutrophil recovery. The study was designed to assess the non-inferiority of the short treatment compared with the extended treatment regimen, with a non-inferiority margin of 10%. The primary outcome was adjudicated by an independent outcome committee, who were masked to treatment allocation, and was analysed in the intention-to-treat and per-protocol populations. The trial is completed and registered with ClinicalTrials.gov, NCT02149329. Findings: Between Dec 1, 2014, and July 1, 2019, 281 patients were included in the intention-to-treat analysis: 144 (51%) patients were assigned to the short treatment group and 137 (49%) to the extended treatment group. Median age was 59 years (IQR 52–65); 109 (39%) patients were women and 172 (61%) were men; 205 (73%) patients received HSCT. In the intention-to-treat analysis, 28 (19%) of 144 patients in the short treatment group versus 21 (15%) of 137 patients in the extended treatment group had treatment failure (adjusted risk difference [ARD] 4·0% [90% CI –1·7% to 9·7%]; p=0·25). In the per-protocol analysis (n=225), 24 (23%) of 104 patients in the short treatment group and 19 (16%) of 121 patients in the extended treatment group had treatment failure (ARD 7·3% [0·3% to 14·9%]; p=0·11). The most common grade 3–5 infection-related adverse events were mucositis (23 [20%] of 114 adverse events in the short treatment group vs 28 [29%] of 98 adverse events in the extended treatment group), fever of unknown origin (20 [18%] vs 16 [16%] events), and bacteraemia (15 [13%] vs 13 [13%] events). The number of serious adverse events were higher in the short treatment group (23 [16%] of 144 patients) than in the extended treatment group (14 [10%] of 137 patients), due to an increased rate of readmission (17 [12%] patients in the short treatment group vs ten [7%] in the extended treatment group). Death before 30 days after neutrophil recovery occurred in five (3%) participants in the short treatment group: two due to progressive leukaemia, two due to candidaemia, and one due to Enterococcus faecium bacteraemia and drug-induced pneumonitis. One (1%) patient died in the extended treatment group due to candidaemia. None of the deaths were related to carbapenem-sensitive infections. Interpretation: Early discontinuation of carbapenem treatment in patients with febrile neutropenia of unknown origin does not result in increased treatment failure. Our study supports short treatment if patients are afebrile after 3 days of carbapenem treatment. However, because secondary analyses suggested that serious adverse events and all-cause mortality occurred more often in patients who are persistantly febrile the short treatment group, we recommend vigilance for non-susceptible pathogens and early resumption of empirical therapy in patients who are deteriorating. Funding: The Netherlands Organisation for Health Research and Development and Fonds NutsOhra.