The high incidence of infections with coagulase-negative staphylococcus has resulted in a very frequent use of vancomycin (V) in the NICU. Levels of V should be at least adequate to kill bacteria with intermediate resistance(IR), which are increasing in prevalence. In adults, peak V levels can be accurately predicted from trough values, so that measuring trough levels may be sufficient in patients with normal renal function. In neonates, however, we expected that high variability of pharmacokinetics of V and renal function would limit the predictability of peak levels from trough values. We administered V according to the protocol described in NeoFax, which is based on the volume of distribution (Vd) and glomerular filtration rate that are expected for gestational age (GA) and postconceptional age (PCA). We obtained a steady state trough level at the fourth dose and a peak level one hour after the end of a 1-hour infusion. We assumed a single compartment distribution, first order kinetics and completion of V distribution at the time of peak level. We measured V levels by Syva EMIT reaction and plasma serum creatinine concentrations (Pcr) by kinetic Jaffe method using a Hitachi 914 or a Delta CX3. We have obtained 100 paired V levels in 72 patients with a GA of 29.4±4.7 weeks (mean ±SD) and a PCA of 32.9 ±4.5 weeks. Six patients had a high peak and a low trough or vice-versa. The agreement between peak and trough was poor (Cohen Kappa 0.18, p<0.014, 95% CI 0.03-0.33). If trough alone had been obtained, adjusting the interval but not the dose, would have led to peak levels that would have been toxic in 6 patients, and subtherapeutic for IR organisms in 4 patients. Peak level was best predicted(r=0.71, p<0.001, n=78) from dose, trough and PCA using the formula: Peak(μg/ml) = 19.02 + 1.07 dose (mg/kg) + 1.05 trough (μg/ml) - 0.47 PCA(wk). The standard error of the estimate was high (6.8), leading to a 95% CI that was twice the width of the normal range for peak values. The clearance, half-life and the Vd of V had an inter-individual coefficient of variance greater than 50%. The prediction of peak V level did not improve by using Pcr, possibly because of the low precision of Pcr measurements by kinetic Jaffe in neonates, especially for levels <1 mg/dl (Clermont et al, 1986).Within the limits of the methods, peak V levels cannot be reliably predicted from trough values in neonates.