Arsenic presents several unique problems in risk assessment. First, there is no good animal model for arsenic as a carcinogen, although in humans arsenic exposure through inhalation is judged to cause lung cancer and ingested inorganic arsenic is judged to cause skin cancer. Second, detoxification of arsenic through methylation is believed to be important, but the mechanisms and the quantitative relationships are not yet understood.EPA provided a risk assessment for ingested inorganic arsenic in its 1984 Health Assessment Document and a revised version in 1988. In both cases EPA calculated a cancer potency or slope factor using epidemiological data from Taiwan. EPA's standard or default risk assessment procedure is to use the linear coefficient from the multistage model in order to calculate cancer risk. This procedure was challenged by the EPA Science Advisory Board (SAB) in a report to the Administrator in September of 1989. The SAB recommended that EPA "(1) develop a revised risk assessment based on estimates of the delivered dose of non-detoxified arsenic to target tissues, and (2) consider the potential reduction in cancer risk due to detoxification in establishing an MCL for arsenic".This paper will draw upon the author's experience with the SAB to summarise major issues in arsenic risk assessment and to examine how these issues might be resolved through further research.