We evaluated the reliability of serum concentrations of procalcitonin for the diagnosis of early- and late-onset sepsis in a neonatal intensive care unit (NICU) setting. Timed procalcitonin determinations were prospectively obtained during two postnatal periods: 0-48 hours of age (period 1) and 3-30 days of age (period 2). In period 1, we measured procalcitonin concentrations in 83 healthy newborns (group 0) and in 120 NICU patients (14 with culture-proven sepsis, group 1A; 14 with clinical septicemia, group 1B; 75 with no evidence of infection, group 2; and 17 with uncertain findings, group 3). After we established 95% hour-specific reference ranges for group 0, we performed multiple linear regression analyses to determine which maternal, intrapartum, and neonatal complications would affect normal procalcitonin values. Maternal diabetes was the only variable identified in group 2 patients that induced a significant deviation from procalcitonin reference ranges. Analyses of the pooled procalcitonin values obtained for group 1 patients over the 48-hour period after birth yielded a sensitivity of 92.6% and a specificity of 97.5% for procalcitonin concentrations in the detection of early-onset sepsis. In period 2, blood samples from 23 cases with systemic infections were analyzed for procalcitonin concentrations at the onset of signs of infection. The control group was formed by matching four uninfected NICU patients to each infected case. None of the procalcitonin values for the 92 controls overlapped those for the cases (sensitivity and specificity, 100%). Procalcitonin is a promising marker for the diagnosis of early- and late-onset sepsis in neonates at high risk for this infection.