The study aims to examine whether the annual variations in fasting plasma glucose (FPG) measurements, represented by the coefficient of variation (CV), predict cancer incidence and mortality in the subsequent years independent of traditional risk factors of type 2 diabetic patients. A computerized database of patients with type 2 diabetes of 30 years old and older (n=4805) enrolled in the Diabetes Care Management Program of a medical center before 2006 was analyzed using a time-dependent Cox's proportional hazards regression model. The mortality rates for the first, second, and third tertiles of the first annual FPG-CV were 8.64, 12.71, and 30.82 per 1000 person-years respectively. After adjusting for mean FPG, HbA1c, and other risk factors, the annual FPG-CV was independently associated with cancer incidence, cancer mortality, and cancer incidence or mortality, and the corresponding hazard ratios for the third vs first tertile of the annual FPG-CV were 3.03 (1.98, 4.65), 5.04 (2.32, 10.94), and 2.86 (1.91, 4.29) respectively. The annual variation in FPG was a strong predictor of cancer incidence and mortality in type 2 diabetic patients; therefore, glucose variation may be important in the clinical practice of care management and cancer prevention.