Indicine N-oxide (INO) is a pyrrolizidine alkaloid (PA) with antitumor activity in animals and humans. Prior studies showed that despite the known hepatic toxicity of the PAs, INO did not produce hepatic toxicity in animals but caused unpredictable lethal hepatic toxicity in humans. In this study we have attempted to find a model system for predicting the hepatotoxic potential of antitumor PAs. Primary cultures of rat hepatocytes showed toxicity only with the most hepatotoxic PAs such as lasiocarpine, but did not detect toxicity with other PAs. Subchronic intraperitoneal administration of PAs to weanling rats and adult mice produced, in surviving animals, hepatic megalocytosis and centrilobular necrosis with heliotrine (H) and 9-O-(R(-)-2-(4'-chlorophenyl)-2-hydroxybutyryl)retronecine N-oxide (RC1NO) but only megalocytosis with INO. Thus, despite previous reports, weanling rats offered no advantage over adult mice for detecting significant hepatic toxicity with PAs. Phenobarbital pretreatment of the mice did not increase the hepatic toxicity of any of the PAs. Subchronic oral administration of PAs to adult mice produced hepatic megalocytosis and centrilobular necrosis in surviving animals with H and RC1NO and megalocytosis with INO. Animals that died acutely following oral administration of INO showed hepatic centrilobular necrosis. Administration of several courses of INO intravenously to dogs produced histological evidence of centrilobular hemorrhagic necrosis. It is concluded that there is no single animal model that will predict hepatic toxicity of the type seen in humans with the antitumor PAs. A combination of studies using adult mice and dogs and lethal doses of the PAs offers the best way of detecting potential hepatic toxicity.