RATIONALE: The duration of antibiotic therapy in critically ill patients with sepsis can result in antibiotic overuse, increasing the risk of developing bacterial resistance. OBJECTIVES: To test the hypothesis that an algorithm based on serial measurements of procalcitonin (PCT) allows reduction in the duration of antibiotic therapy compared with empirical rules, and does not result in more adverse outcomes in patients with severe sepsis and septic shock. METHODS: In patients randomly assigned to the intervention group, antibiotics were stopped when PCT levels had decreased 90% or more from the initial value (if clinicians agreed) but not before Day 3 (if baseline PCT levels were <1 microg/L) or Day 5 (if baseline PCT levels were >/=1 microg/L). In control patients, clinicians decided on the duration of antibiotic therapy based on empirical rules. MEASUREMENTS AND MAIN RESULTS: Patients assigned to the PCT group had 3.5-day shorter median duration of antibiotic therapy for the first episode of infection than control subjects (intention-to-treat, n = 79, P = 0.15). In patients in whom a decision could be taken based on serial PCT measurements, PCT guidance resulted in a 4-day reduction in the duration of antibiotic therapy (per protocol, n = 68, P = 0.003) and a smaller overall antibiotic exposure (P = 0.0002). A similar mortality and recurrence of the primary infection were observed in PCT and control groups. A 2-day shorter intensive care unit stay was also observed in patients assigned to the PCT group (P = 0.03). CONCLUSIONS: Our results suggest that a protocol based on serial PCT measurement allows reducing antibiotic treatment duration and exposure in patients with severe sepsis and septic shock without apparent harm.