Aim To investigate the appropriate fasting plasma glucose threshold by which to define prediabetes in children and adolescents, based on its ability to predict incident paediatric diabetes.Methods In a nationwide survey of diabetes and renal disease conducted between 1992 and 2000 in all school-aged children in Taiwan, those with abnormal results in repeated urine tests received further physical examination and blood tests. Students who had blood tests for at least two time points were selected for the present study (N = 12 119). The incidence of paediatric diabetes, adjusted hazard ratio and predictive power of fasting plasma glucose were analysed.Results The incidence of paediatric diabetes increased with increasing fasting plasma glucose levels. Groups with fasting plasma glucose >5.6 mmol/l had a higher adjusted hazard ratio. The adjusted hazard ratio of incident diabetes for participants with higher fasting plasma glucose rose continuously when using a higher threshold for fasting plasma glucose. The area under the receiver-operating characteristic curve for fasting plasma glucose was 0.628 for predicting paediatric diabetes. The association between fasting plasma glucose and incident paediatric diabetes and the area under the receiveroperating characteristic curve were similar in boys and girls and were higher in the age group 12-18 years. According to receiver-operating characteristic curve analysis, the optimal thresholds, sensitivity and specificity were 4.75 mmol/l, 65% and 51%, respectively, for those aged 6-11 years and 5.19 mmol/l, 60% and 73%, respectively, for those aged 12-18 years.Conclusion Fasting plasma glucose is associated with the incidence of paediatric diabetes. The results of the present study can be used as reference data to suggest a cut-off value to define paediatric prediabetes.