This study tested the hypothesis that the determinants of mild liver injury are prerequisites for more severe idiosyncratic hepatotoxicity. This study verified whether the possible risk factors for rare idiosyncratic valproic acid (VPA)-induced hepatotoxicity, VPA clearance and/or serum carnitine concentrations are common to those for a mild elevation in transaminases in VPA-treated patients. VPA clearance was calculated in 172 Japanese patients with epilepsy, using a non-linear mixed-effects regression program. Carnitine concentrations were determined in a subset of 60 patients. The relationships between VPA clearance, carnitine concentration and levels of transaminases and ammonia were evaluated by Pearson's correlation coefficients. The final model of VPA apparent clearance (CL/F) was as follows: CL/F (L h(-1) = 0.012 x (BW/40)(0.34) x dose(0.55) x 0.90(gender) x 1.32(PHT) x 1.11(CBZ) x 1.12(PB), where BW = total body weight (kg); gender = 1 if female, 0 if male; PHT/CBZ/PB = 1 if phenytoin, carbamazepine, or phenobarbital, respectively, is coadministrated, otherwise 0. Either a higher VPA clearance or acyl/free carnitine ratio and a lower total and/or free carnitine concentration, but not VPA concentration, were associated with the mild elevation in transaminases or ammonia. These results support the initial hypothesis, while also helping to clarify the mechanism of severe idiosyncratic hepatotoxicity with VPA.