INTRODUCTION: With diabetes defined by HbA1c≥6.5% and/or FPG≥7.0mmol/l and/or diabetes treatment, we investigated HbA1c and fasting plasma glucose (FPG) thresholds/change-points above which the incidence of diabetes increases. METHODS: Data are Danish (Inter99), Australian (AusDiab) and French (D.E.S.I.R.), with respectively 4930, 6012 and 3784 non-diabetic participants. RESULTS: Diabetes incidences at 5 years for Inter99 and AusDiab and at 6 years for D.E.S.I.R. were 2.3%, 3.1% and 2.4% respectively and incidences increased with baseline HbA1c and FPG. As HbA1c distributions differed between cohorts, HbA1c was standardized on D.E.S.I.R. data. Change-points where diabetes incidence increased were identified for HbA1c (%) after standardization: 5.1 (4.9-5.6) (Inter99), 5.4 (5.1-5.6) (AusDiab), 5.3 (5.1-5.7) (D.E.S.I.R.); for FPG change-points (mmol/l) were 5.1 (...-6.1) (Inter99), 5.5 (5.2-5.8) (AusDiab), no change-point for D.E.S.I.R. Using current diabetes risk criteria HbA1c≥5.7% and/or FPG≥5.6mmol/l to screen for diabetes provided high sensitivity (over 89%) and positive predictive values: 4.3%, 6.9%, and 5.9% respectively. CONCLUSIONS: HbA1c and FPG change-points predicting incident diabetes did not always exist, differed across studies, when available were generally lower than current criteria with wide confidence intervals. Using jointly HbA1c≥5.7% and/or FPG≥5.6mmol/l as a criterion for the risk of incident diabetes is appropriate.