OBJECTIVE To find out whether a newly found association between diabetes mellitus and arsenic in drinking water in Taiwan could be reproduced in copper smelters with arsenic exposure. METHODS Extended analysis of a previous case-control study from 1978 was based on death records and objective exposure information from the company. The final analysis included only those employed at the smelter. Cases were 12 people with diabetes mellitus on the death certificate and those for whom there was clinical information on this disease. Controls were 31 people without cancer, cardiovascular and cerebrovascular disease as these disease categories had been associated with arsenic exposure in the original study and elsewhere. RESULTS The odds ratios found for diabetes mellitus with increasing arsenic exposure categories were (reference level = 1), 2.0, 4.2, and 7.0, but the 95% confidence intervals included unity. Unstratified test for trend was weakly significant, P = 0.03. CONCLUSIONS Although based on small numbers, the findings provide some support for the suggestion that arsenic exposure could sometimes play a part in the development of diabetes mellitus.