BACKGROUND To evaluate the relationship between hemoglobin A1c variability and all-cause mortality in type 2 diabetic patients. METHODS This was a retrospective cohort study in type 2 diabetic patients followed for at least 2 years between 2003 and 2009. A1C variability was determined from the standard deviation or coefficient of variation of serial A1C values (A1C(SD) or A1C(CV)). Subjects were categorized into either the high or low A1C variability group according to their A1C(CV) median. Hazard ratios (HRs) of various factors for all-cause mortality were determined from Cox's proportional hazard models. RESULTS A total of 881 subjects (422 men, 459 women) were included and 73 (8.3%) died during follow-up. The follow-up period was 4.7 ± 2.3 years. All-cause mortality was higher in subjects with high A1C(CV) (11.0% vs. 5.4%, p=0.002). In the Kaplan-Meier failure curve, subjects with higher A1C(CV) demonstrated a trend of higher mortality (p=0.1). In multivariate Cox's proportional hazards models, A1C(SD) and A1C(CV) significantly predicted all-cause mortality with an HR of 1.987 (p=0.02) and 1.062 (p=0.013), respectively, after adjusting for age, gender, body mass index, duration of diabetes, mean systolic blood pressure, use of antihypertensives and statins, mean LDL-cholesterol, smoking status, chronic kidney disease, and mean A1C values (A1C(MEAN)). The ability of A1C(SD) and A1C(CV) to predict all-cause mortality was more evident in subjects with relatively low A1C(MEAN.) CONCLUSIONS A1C variability is an important risk factor for all-cause mortality in type 2 diabetic patients.