The effects of smoking on the 8-year (1984-1992) incidence of non- insulin-dependent diabetes mellitus (NIDDM) were investigated in a cohort of 2,312 male employees of an electrical company in Japan. The incidence rate was 2.2/1,000 person-years. After controlling for other known risk factors for NIDDM, a proportional hazards regression analysis indicated that those who were currently smoking 16-25 cigarettes per day had a 3.27 times higher risk of developing NIDDM during the follow-up period than never smokers (p < 0.05); the hazard ratio was similar (3.21) for those who were currently smoking ≤26 cigarettes per day. Among ever smokers, proportional hazards regression analysis also indicated that younger age at starting smoking was associated with an increased risk of NIDDM (p for trend = 0.09). These findings suggest that number of cigarettes smoked per day is important in the relation between smoking and NIDDM incidence, although the risks did not increase above a given amount of smoking. Starting smoking at a younger age also might be independently associated with an increased risk of NIDDM.