To explore the relation between respiratory cancer mortality and exposure to airborne arsenic, methods of conditional logistic regression analysis are applied to a matched case-control study of copper smelter employees in Montana. With follow-up data for the period 1938-1977, several measures of arsenic exposure are compared. For men first employed prior to 1925, three measures--the category of maximum arsenic exposure, cumulative arsenic exposure, and time-weighted average arsenic exposure--are good predictors of respiratory cancer mortality. Within this group, there was a significantly elevated relative risk of respiratory cancer mortality associated with medium and heavy exposure to arsenic. For men first employed during 1925-1947, time-weighted average exposure to arsenic (adjusted for age at first employment) is the best predictor. In this group, men with heavy arsenic exposure and initially employed at 16.9 years had a relative risk of 6.0 in comparison to the baseline group with only light exposure to arsenic and initially employed at 31.9 years. No particular advantage is found in lagging exposures 10 years prior to death of the case in each matched set.