Aim: We examined the ability of fasting plasma glucose and HbA 1c to predict 5-year incident diabetes for an Australian cohort and a Danish cohort and 6-year incident diabetes for a French cohort, as defined by the corresponding criteria. Methods We studied 6025 men and women from AusDiab (Australian), 4703 from Inter99 (Danish) and 3784 from DESIR (French), not treated for diabetes and with fasting plasma glucose <7.0mmol/l and HbA 1c<48mmol/mol (6.5%) at inclusion. Diabetes was defined as fasting plasma glucose ≥7.0mmol/l and/or treatment for diabetes or as HbA 1c≥48mmol/mol (6.5%) and/or treatment for diabetes. Results For AusDiab, incident fasting plasma glucose-defined diabetes was more frequent than HbA 1c-defined diabetes (P McNemar<0.0001), the reverse applied to Inter99 (P McNemar<0.007) and for DESIR there was no difference (P McNema=0.17). Less than one third of the incident cases were detected by both criteria. Logistic regression models showed that baseline fasting plasma glucose and baseline HbA 1c predicted incident diabetes defined by the corresponding criteria. The standardized odds ratios (95% confidence interval) for HbA 1c were a little higher than for fasting plasma glucose, but not significantly so. They were respectively, 5.0 (4.1-6.1) and 4.1 (3.5-4.9) for AusDiab, 5.0 (3.6-6.8) and 4.8 (3.6-6.3) for Inter99, 4.8 (3.6-6.5) and 4.6 (3.6-5.9) for DESIR. Conclusions Fasting plasma glucose and HbA 1c are good predictors of incident diabetes defined by the corresponding criteria. Despite Diabetes Control and Complications Trial-alignment of the three HbA 1c assays, there was a large difference in the HbA 1c distributions between these studies, conducted some 10years ago. Thus, it is difficult to compare absolute values of diabetes prevalence and incidence based on HbA 1c measurements from that time.