The current standard approach for assessing carcinogenic potential is to conduct a near lifetime rodent pathology study with the high dose set to the maximum tolerated dose (MTD) of the test chemical. The linearized multistage model is then used as the default approach to estimate the potential human cancer risk at environmental elvels of the chemical. There is an increasing appreciation in the scientific and regulatory communities that chemical carcinogens differ dramatically in potency, exhibit a high degree of tissue and species specificity, and act through different modes of action. This paper advocates a decision tree strategy for classifying carcinogens that are acting primarily through genotoxic, cytotoxic, or mitogenic pathways. A primary concern is whether the chemical has direct genotoxic potential resulting from DNA reactivity or clastogenicity of the compound or its metabolite(s). Knowledge of the exposure-response curve for cytotoxicity is important because initiation and promotion events may occur secondary to a variety of associated activities such as regenerative cell proliferation. Mitogens indice direct stimulation of growth and may provide a selective growth advantage to spontaneously initiated precancerous cells. Of particular concern is the situation where pathological changes induced during the course of the treatment at high doses near the MTD are absent at lower, environmentally relevant, doses. If the tumor response is coincident with the preceding toxic response, it may not be justified to use the high-dose data in extrapolating to expected responses at low environmental exposures where no induced tissue abnormalities occur. Suggestions are presented for appropriate risk assessment approaches for different modes of action. Examples discussed are formaldehyde, a weakly genotoxic rodent nasal carcinogen; chloroform, a nongenotoxic-cytotoxic rodent liver and kidney carcinogen; and phenobarbital, a nongenotoxic-mitogenic rodent liver carcinogen.