This communication describes a modification of the total plasma phenytoin enzyme immunoassay technique (EMIT) run on the Cobas MIRA analyser that allows quantitation of unbound phenytoin concentrations in human plasma for routine therapeutic drug monitoring (TDM) purposes. An application of this method is also presented to consider the previously described protein binding drug interaction with concomitantly administered valproic acid in patients with epilepsy. The coefficients of variation for the unbound phenytoin assay ranged from 7.5 to 9.6% and the assay had a reproducibility and accuracy similar to the total phenytoin assay, acceptable for routine TDM. Phenytoin protein binding was linear over a range of total plasma concentrations of 3–65 mg/L. Patients also receiving valproic acid (nine patients, 105 specimens) had a significantly (p< 0.0001) greater mean ± SD unbound phenytoin fraction (13.3 ± 3.1%) compared to nine patients (110 specimens) not receiving valproic acid (8.3 ± 1.6%). There was also a significant correlation (p< 0.001) between plasma valproic acid concentration and unbound phenytoin fraction, which resulted in greater intrasubject variability in phenytoin protein binding.