We examined the reproducibility of oral glucose tolerance tests (OGTT) using the World Health Organization criterion in 212 Chinese subjects (male 149, female 63) who underwent two 75 g OGTTs within a 6-week period. The overall reproducibility was 65.6% (139/212) of which 74 subjects had normal glucose tolerance, 24 had diabetes and 41 had impaired glucose tolerance (IGT) on two occasions. The subjects were divided into three groups [group 1: normal OGTTs on both occasions (n = 74); group 2: one abnormal OGTT (either diabetes or IGT (n = 51); group 3: 2 abnormal OGTTs (n = 87)]. Subjects in group 1 were younger, had lower blood pressure, body mass index (BMI), waist-to-hip ratio (WHR), fasting and 2 h plasma insulin levels, triglyceride, very-low density lipoprotein and apolipoprotein-B concentrations than both groups 2 and 3. Group 2 had similar characteristics as group 3 except for a lower glycated haemoglobin (HbA1c), fasting and 2 h plasma glucose during the two OGTTs. With receiver operating characteristic curve (ROC) analysis, a HbA1c of 5.3% gave an optimal sensitivity of 70.7% and specificity of 74.3% to predict diabetes as defined by a 2 h plasma glucose value > or = 11.1 mmol/L in the first OGTT. Of the 212 subjects, 73 had HbA1c > or = 5.3%. The reproducibility of OGTT was 56.2% for these 73 subjects. With ROC analysis, a BMI of 25 kg/m2 gave an optimal sensitivity of 53.7% and specificity of 56.7% to predict diabetes. For the 36 subjects with BMI > or = 25 kg/m2, the reproducibility of OGTT was 58.3%. Similarly, for the 140 subjects with WHR > or = 0.9, the reproducibility of OGTT was 57.9%. These findings confirmed the poor reproducibility of OGTT which was not improved even amongst subjects with high HbA1c, BMI or WHR. Furthermore, subjects with one abnormal OGTT, whether reproducible or not, had a higher cardiovascular risk profile compared to subjects who had two normal OGTTs.