Because of the potential of arsenic for causing cancer in humans, and of the fact of widespread environmental and occupational exposure, deriving acceptable human-limit values has been of major concern to industry as well as to regulatory agencies. Based upon epidemiological evidence and mechanistic studies, it has been argued that a non-linear dose-response model at low-level exposures is more appropriate for calculating risk than the more commonly employed linear-response models. In the present studies, dose-response relationships and recovery studies employing a cancer precursor marker, i.e., activating protein (AP)-1 DNA-binding activity, were examined in bladders of mice exposed to arsenic in drinking water and compared to histopathological changes and arsenic tissue levels in the same tissue. While AP-1 is a functionally pleomorphic transcription factor regulating diverse gene activities, numerous studies have indicated that activation of the MAP kinase pathway and subsequently increased AP-1 binding activities, is a precursor for arsenic-induced cancers of internal organs as well as the skin. We observed previously that within 8 weeks of exposure AP-1 activation occurs in urinary bladder tissue of mice exposed to arsenic in the drinking water. In the present studies, C57BL/6 mice were exposed to sodium arsenite at various concentrations in the drinking water for 8 consecutive weeks. Minimal but observable AP-1 activity occurred in bladder tissue at exposure levels below which histopathological changes or arsenic tissue accumulation was detected. Marked AP-1 DNA-binding activity only occurred at exposure levels of sodium arsenite above 20 microg/ml, where histopathological changes and accumulation of arsenic in the urinary bladder epithelium occurred. Although the experimental design did not allow statistical modeling of the entire dose-response curve, the general shape of the dose-response curve is not inconsistent with the previously proposed hypothesis that arsenic-induced cancer follows a non-linear dose-response model.