Limited performance of the octanol-water partition coefficient has been known in predicting the brain/blood equilibrium distribution ratio. There have been reports that predictive ability of the in vitro partition parameters increases after the introduction to regression of a correction term reflecting bulkiness of the compounds. Here a rationalization for such observations has been proposed by considering the brain/blood and the octanol-water distribution as composed of a part determined by polarity of a compound and another part determined by its nonspecific, dispersive properties. The model proposed was shown theoretically to apply to the four available representative sets of brain/blood distribution data. Instead of a search for an in vitro partition system precisely mimicking the brain/blood distribution equilibrium a model is recommended by a combination of standard partition parameters with a molecular bulkiness descriptor.