Current approaches to risk assessment typically assume a linear dose-response for mutagenic compounds that directly interact with DNA or when the carcinogenic mechanism is unknown. Because the mode of action of arsenic-induced carcinogenesis is not well established, recent dose-response assessments for arsenic have assumed linearity at low doses despite evidence that arsenic is not a direct-acting mutagen. Several modes of action, including generation of oxidative stress, perturbation of DNA methylation patterns, inhibition of DNA repair, and modulation of signal transduction pathways, have been proposed to characterize arsenic's toxicity. It is probable that these mechanisms do not act in isolation, but overlap, and contribute to the complex nature of arsenic-induced carcinogenesis. All of the proposed mechanisms are likely to be nonlinear at low does. Furthermore, studies of populations outside the US exposed to arsenic in drinking water show increases in cancer only at relatively high concentrations, that is, concentrations in drinking water of several hundred micrograms per liter (microg/l). Studies in the US of populations exposed to average concentrations in drinking water up to about 190 microg/l do not provide evidence of increased cancer. Consideration of arsenic's plausible mechanisms and evidence from epidemiological studies support the use of nonlinear methods, either via biologically based modeling or use of a margin-of-exposure analysis, to characterize arsenic risks.