It has been proposed that initial empirical broad-spectrum antibiotic therapy will result in better clinical outcomes and that shorter courses will reduce the 'collateral damage' of promoting antibiotic resistance. There are few data from Intensive Care Units (ICUs) that support this latter conclusion. A prospective observational study was undertaken at the National University Hospital, Singapore, to examine the relationship between duration of carbapenem therapy and subsequent nosocomial multidrug-resistant (MDR) bloodstream infection (BSI). Over a 2-year period, 415 ICU patients receiving empirical carbapenem therapy were prospectively followed. MDR BSI occurred on 35 occasions in 31 patients, comprising 21 carbapenem-resistant Acinetobacter baumannii, 3 carbapenem-resistant Pseudomonas aeruginosa and 11 meticillin-resistant Staphylococcus aureus (MRSA). There was no difference in the duration of carbapenems for those who developed MDR BSI compared with those who did not [median duration 8 days (range 3-23 days) vs. 9 days (range 3-59 days); P=0.78]. On multivariate analysis using the Cox proportional hazard model the hazard ratio was 0.935 (P=0.070). In this cohort of critically ill patients, a shorter duration of carbapenem therapy was not shown to protect against subsequent development of MDR BSI. Strategies that depend primarily on reducing broad-spectrum antibiotic duration may be inadequate in preventing the emergence of MDR organisms.