A battery of in vitro methods has been developed for the prediction of acute oral toxicity, to reduce the number of animals used for this purpose. However, the results of these tests correlate more closely with lethal serum concentrations than with lethal doses. To address this issue, we have further evaluated the HPCT-1E3 model, which may be better able to emulate toxicokinetic factors that occur in vivo, due to the presence in these hepatocytoma cells of endogenous transmembrane carriers and a basal activity of xenobiotic metabolism. IC50 values produced by using the MTT test after a 48-hour incubation with 20 randomly-selected MEIC substances, correlated better with human oral LD50 values than with LC50 data, supporting this hypothesis. As with other models, the toxicity of receptor-specific rather than cytotoxic substances, for example digoxin, was underpredicted. When digoxin was removed from the correlation analysis, the coefficient of determination (r(2)) improved to 0.81, and none of remaining chemicals were wrongly predicted by more than one order of magnitude. IC50 values obtained with HepG2 cells under similar conditions (MEIC Test No. 3, 24 hours, MTT) correlated with human LD50 data with a r(2) value of 0.55. A direct comparison of HPCT-1E3 and HepG2 cells further suggested that the differences between them may be due to transport processes. In conclusion, the HPCT-1E3 model may be valuable in improving the prediction of lethal doses, rather than lethal serum concentrations.