Dosing of phenytoin is difficult in children because of its variable pharmacokinetics and protein binding. Possible covariates for this protein binding have mostly been univariately investigated in small, and often adult, adult populations. We conducted a study to identify and quantify these covariates in children. We extracted data on serum phenytoin concentrations, albumin, triglycerides, urea, total bilirubin and creatinine concentrations and data on coadministration of valproic acid or carbamazepine in 186 children. Using nonlinear mixed effects modeling the effects of covariates on the unbound phenytoin fraction were investigated. Serum albumin, serum urea concentrations, and concomitant valproic acid use significantly influenced the unbound phenytoin fraction. For clinical practice, we recommend that unbound phenytoin concentrations are measured routinely. However, if this is impossible, we suggest to use our model to calculate the unbound concentration. In selected children, close treatment monitoring and dose reductions should be considered to prevent toxicity.