Background Whether antibiotic rotation strategies reduce prevalence of antibiotic-resistant, Gram-negative bacteria in intensive care units (ICUs) has not been accurately established. We aimed to assess whether cycling of antibiotics compared with a mixing strategy (changing antibiotic to an alternative class for each consecutive patient) would reduce the prevalence of antibiotic-resistant, Gram-negative bacteria in European intensive care units (ICUs). Methods In a cluster-randomised crossover study, we randomly assigned ICUs to use one of three antibiotic groups (third-generation or fourth-generation cephalosporins, piperacillin–tazobactam, and carbapenems) as preferred empirical treatment during 6-week periods (cycling) or to change preference after every consecutively treated patient (mixing). Computer-based randomisation of intervention and rotated antibiotic sequence was done centrally. Cycling or mixing was applied for 9 months; then, following a washout period, the alternative strategy was implemented. We defined antibiotic-resistant, Gram-negative bacteria as Enterobacteriaceae with extended-spectrum β-lactamase production or piperacillin–tazobactam resistance, and Acinetobacter spp and Pseudomonas aeruginosa with piperacillin– tazobactam or carbapenem resistance. Data were collected for all admissions during the study. The primary endpoint was average, unit-wide, monthly point prevalence of antibiotic-resistant, Gram-negative bacteria in respiratory and perineal swabs with adjustment for potential confounders. This trial is registered with ClinicalTrials.gov, number NCT01293071. Findings Eight ICUs (from Belgium, France, Germany, Portugal, and Slovenia) were randomly assigned and patients enrolled from June 27, 2011, to Feb 16, 2014. 4069 patients were admitted during the cycling periods in total and 4707 were admitted during the mixing periods. Of these, 745 patients during cycling and 853 patients during mixing were present during the monthly point-prevalence surveys, and were included in the main analysis. Mean prevalence of the composite primary endpoint was 23% (168/745) during cycling and 22% (184/853) during mixing (p=0·64), yielding an adjusted incidence rate ratio during mixing of 1·039 (95% CI 0·837–1·291; p=0·73). There was no difference in all-cause in-ICU mortality between intervention periods. Interpretation Antibiotic cycling does not reduce the prevalence of carriage of antibiotic-resistant, Gram-negative bacteria in patients admitted to the ICU.