OBJECTIVES: The association of A1C’s variability with developing new diabetes has been little studied. We aimed to evaluate the effect of visit-to-visit A1C variability on the risk of developing new diabetes in healthy adults in Japan. METHODS: Population-based, cohort study from 2005 to 2008 in Tokyo, Japan of healthy adults not taking diabetes medication and with a HbA1c lower than 6.5% at baseline. Based on annual measurement of serum HbA1c we calculated the annual visit-to-visit variability, and used this as a predictor of new onset diabetes in a multivariate logistic regression. RESULTS: At baseline, 14,587 people (50% female) with a mean age of 51 years old (SD: 12 years, range: 23 to 92), a mean fasting plasma glucose (FPG) level of 98.4 mg/dl (SD: 9.3 mg/dl) and a mean HbA1c level of 5.3 % (SD: 0.4 %) had annual check-ups over 4 years. After adjusting for the other potential risk factors new diabetes was predicted by the A1C variability (odds ratio (OR): 10.3 for highest (>= 0.16%)) versus the lowest quantile (<0.08 %), 95%CI: 5.9 – 18.0) and by the baseline A1C (OR: 55.2 for A1C of 6.0 – 6.4 % versus A1C of <5.0 %, 95% CI: 13.2 – 230). FPG (OR: 1.1, 95%CI: 1.1 – 1.2) and Smoker (OR: 1.8, 95%CO: 1.3 – 2.6) weakly but also significantly related to develop the new diabetes. For predicting the development of diabetes, the combination of the level of AIC at baseline and the variability (AUC for the ROC=0.94) was superior to the level of A1C at baseline alone (AUC=0.89). CONCLUSIONS: Visit-to-visit variability in A1C independently added to the baseline A1C in predicting the risk of developing new diabetes for the healthy adults. We should consider not only the baseline A1c level but also variability in A1C to prevent development of the diabetes.