Risk of weight change and fluctuation for type 2 diabetes was studied in a prospective study among 20,952 Finnish male smokers aged 50-69 years. Baseline data on lifestyle and medical history were collected, and height and weight measured. Weight was measured thrice after the baseline, 1 year apart. Weight change was defined as the slope of the regression line fitted to the four measurements and weight fluctuation as the root-mean-square-error deviation from this line. Incident cases of diabetes were identified from a national medication reimbursement register; 535 cases up to 9 years' follow-up. The Cox proportional hazards model served to estimate relative risk [RR, 95% confidence interval (CI)]. Weight gain and fluctuation associated with higher risk for diabetes, multivariate RR = 1.77, 95% CI 1.44-2.17, for weight gain of at least 4 kg compared with those of weight change less than 4 kg, and RR = 1.64, 95% CI 1.24-2.17 in the highest weight fluctuation quintile compared to the lowest. These RRs remained similar when weight change and fluctuation were adjusted for each other. Large weight fluctuation increased the risk of diabetes both in men who gained weight (>or=4 kg), had stable weight (+/- <4 kg), and lost weight (>or=4 kg); RR = 2.17, 95% CI 1.60-2.94, RR = 1.47, 95% CI 1.14-1.91, and RR = 2.04, 95% CI 1.47-2.83, respectively, compared to those with stable weight and moderate fluctuation. Short-term weight gain and large weight fluctuation are independent risk factors for diabetes.