Introduction Few studies have addressed the decision-making process of antibiotic therapy (AT) in intensive care unit (ICU) patients. Methods In a prospective observational study, all consecutive patients admitted over a one-month period (2004) to 41 French surgical (n = 22) or medical/medico-surgical ICUs (n = 19) in 29 teaching university and 12 non-teaching hospitals were screened daily for AT until ICU discharge. We assessed the modalities of initiating AT, reasons for changes and factors associated with in ICU mortality including a specific analysis of a new AT administered on suspicion of a new infection. Results A total of 1,043 patients (61% of the cohort) received antibiotics during their ICU stay. Thirty percent (509) of them received new AT mostly for suspected diagnosis of pneumonia (47%), bacteremia (24%), or intra-abdominal (21%) infections. New AT was prescribed on day shifts (45%) and out-of-hours (55%), mainly by a single senior physician (78%) or by a team decision (17%). This new AT was mainly started at the time of suspicion of infection (71%) and on the results of Gram-stained direct examination (21%). Susceptibility testing was performed in 261 (51%) patients with a new AT. This new AT was judged inappropriate in 58 of these 261 (22%) patients. In ICUs with written protocols for empiric AT (n = 25), new AT prescribed before the availability of culture results (P = 0.003) and out-of-hours (P = 0.04) was more frequently observed than in ICUs without protocols but the appropriateness of AT was not different. In multivariate analysis, the predictive factors of mortality for patients with new AT were absence of protocols for empiric AT (adjusted odds ratio (OR) = 1.64, 95% confidence interval (95%CI): 1.01 to 2.69), age ≥60 (OR = 1.97, 95% CI: 1.19 to 3.26), SAPS II score >38 (OR = 2.78, 95% CI: 1.60 to 4.84), rapidly fatal underlying diseases (OR = 2.91, 95% CI: 1.52 to 5.56), SOFA score ≥6 (OR = 4.48, 95% CI: 2.46 to 8.18). Conclusions More than 60% of patients received AT during their ICU stay. Half of them received new AT, frequently initiated out-of-hours. In ICUs with written protocols, empiric AT was initiated more rapidly at the time of suspicion of infection and out-of-hours. These results encourage the establishment of local recommendations for empiric AT.