Steady-state serum levels of total and unbound valproic acid as well as unbound fraction in epileptic children were studied in a clinical setting. Valproic acid binding parameters were analyzed and compared with in vitro findings. Daily dose of sodium valproate ranging from 29 to 73 mg/kg/day were administered per os. Considerable variation in total and unbound concentrations and unbound fractions within and between subjects was observed. In subjects evaluated in this study, serum level of total and unbound valproic acid ranged from 279 to 1,196 mumol/L and from 37 to 410 mumol/L, respectively. The unbound fraction ranged from 10.32 to 48.39%. In vivo binding parameters obtained from clinical material were as follows: association constant, Ka = 4.984 L/mmol; total binding sites, NP = 1.451 mmol/L, where P is the molar concentration of albumin; number of binding sites per molecule of albumin, N = 2.48. Using spiked sera, binding parameters of Ka = 8.032 L/mmol, NP = 1.262 mmol/L, and N = 1.86 were found in the in vitro study. The association constant obtained from in vivo and in vitro studies were not significantly different (p greater than 0.05) from each other. The unbound fraction of valproic acid was concentration dependent even within the therapeutic range. An equation for estimating unbound concentration (Cf') or unbound fraction (fp') from total concentration (Ct) of valproic acid is derived. The ratio of observed unbound fraction to the estimated unbound fraction (fp/fp') was used to evaluate the variation in valproate serum binding of that clinical sample. Nine samples from hospitalized patients whose medication and diet were closely supervised showed an fp/fp' ratio very close to 1 (mean +/- SD 1.04 +/- 0.24). It is suggested that a clinical sample showing a value of fp/fp' greater than 1.76 (mean + 3 SD) should be evaluated for the cause of the decrease in serum binding and for the associated pharmacokinetic alterations. Therefore, in clinical monitoring of valproate, determination of both total and unbound drug levels was preferable to determination of either one alone. Furthermore, an understanding of the unbound fraction of valproic acid would significantly contribute to the effective management of epileptic patients.