Background Nosocomial infections (NIs) are a leading cause of mortality and morbidity in premature infants. We present a new method for detecting and confirming NIs in a neonatal intensive care unit. Methods Newborns with birth weight < 1,500 g or gestational age (GA) < 33 weeks were included prospectively over 2 years in a single-center tertiary neonatal intensive care unit. The computerized physician order entry system (CPOE) generated alerts when antibiotics were prescribed for at least 5 consecutive days and these cases were reviewed by an expert group following international recommendations. Results Four hundred sixty-one neonates were included, with a mean GA of 30 weeks (range, 26-32 weeks) and mean birth weight 1,270 g (range, 950-1600 g). The CPOE flagged 158 cases of potential NI, 85.1% of which were classified as true NI and 14.9% of which were false positive. Incidence and device-associated nosocomial bloodstream infection rates were 21.9% and 10.8 per 1,000 central venous catheter days, respectively. GA ≤ 28 weeks (odds ratio, 2.18; 95% confidence interval, 1.2-4) and > 7 central venous catheter days (odds ratio, 1.47; 95% confidence interval, 1.3-1.7) were independently associated with the risk of nosocomial bloodstream infection. Conclusion Combining CPOE and interdisciplinary review may improve the accuracy of NI recording in a neonatal intensive care unit.