Abstract Coagulase-negative Staphylococcus has been identified as the main nosocomial agent of neonatal late-onset sepsis. However, based on the pharmacokinetics and erratic distribution of vancomycin, recommended empirical dose is not ideal, due to the inappropriate serum levels that have been measured in neonates. The aim of this study was to evaluate serum levels of vancomycin used in newborns and compare the prediction of adequate serum levels based on doses calculated according to mg/kg/day and m2/day. This is an observational reprospective cohort at a referral neonatal unit, from 2011 to 2013. Newborns treated with vancomycin for the first episode of late-onset sepsis were included. Total dose in mg/kg/day, dose/m2/day, age, weight, body surface and gestational age were identified as independent variables. For predictive analysis of adequate serum levels, multiple linear regressions were performed. The Receiver Operating Characteristic curve for proper serum vancomycin levels was also obtained. A total of 98 patients received 169 serum dosages of the drug, 41 (24.3%) of the doses had serum levels that were defined as appropriate. Doses prescribed in mg/kg/day and dose/m2/day predicted serum levels in only 9% and 4% of cases, respectively. Statistical significance was observed with higher doses when the serum levels were considered as appropriate (p < 0.001). A dose of 27 mg/kg/day had a sensitivity of 82.9% to achieve correct serum levels of vancomycin. Although vancomycin has erratic serum levels and empirical doses cannot properly predict the target levels, highest doses in mg/kg/day were associated with adequate serum levels.