This paper is a reanalysis of data on the respiratory cancer mortality experience of 2,802 men who worked one year or more during the period 1940-1964 at a copper smelter in Tacoma, Washington. Exposure estimates presented earlier have been recalculated and perhaps improved. While the previous analysis showed only a weak relation between respiratory cancer and arsenic exposure, use of new data shows a much stronger relation--but one that is concave downward and not ordinarily considered for environmental exposure and cancer. This new analysis indicates that arsenic is probably more potent as a carcinogen than indicated by other studies. It also demonstrates the distinction between airborne arsenic and the bioavailability of arsenic, and the importance of this distinction for risk assessment. When a dose-response relation is based on airborne concentrations of arsenic, it is clearly concave downward, but when based on urine concentrations, it appears to be linear.