Aims/hypothesis. To study the pathophysiological importance of changes in insulin sensitivity and islet function over time for alterations in glucose tolerance in a randomly selected large group of non-diabetic women aged 57–59 years over a 3-year period.¶Methods. At baseline and at the 3-year follow-up, glucose tolerance (WHO 75 g oral glucose), insulin sensitivity (euglycaemic, hyperinsulinaemic clamp) and insulin and glucagon secretion (2 to 5-min responses to 5 g i. v. arginine at fasting, 14 and > 25 mmol/l glucose) were measured.¶Results. At baseline, women with impaired glucose tolerance (IGT, n = 28) had lower insulin sensitivity (p = 0.048) than normal women (NGT, n = 58). The arginine-induced insulin responses (AIR) were inversely associated with insulin sensitivity (r≥– 0.55, p < 0.001). When related to the 3-year follow-up, the baseline product of AIR at 14 mmol/l glucose times insulin sensitivity, insulin effect index (IE) (r = – 0.40, p < 0.001) and the arginine-induced glucagon response at 14 mmol/l glucose (AGR, r = 0.28, p = 0.009) both correlated with follow-up 2-h glucose. In a multiple regression model, baseline 2-h glucose, insulin effect index and arginine-induced glucagon response independently predicted 2-h glucose at follow-up (total r = 0.668, p < 0.001). Furthermore, Δinsulin sensitivity (i. e. follow-up minus baseline) correlated with Δinsulin secretion (r = – 0.30, p = 0.006), whereas Δglucagon secretion correlated with Δ2-h glucose (r = 0.30, p = 0.006) over the 3 years. In a multiple regression, alterations in 2-h glucose over the 3 years were independently determined by changes in fasting insulin and glucagon secretion (r = 0.424, p < 0.001).¶Conclusion/interpretation. Low insulin secretion, when judged in relation to insulin sensitivity, and high glucagon secretion, determine glucose tolerance over time in the individual subject. These processes are therefore potential targets for prevention of deterioration in glucose tolerance. [Diabetologia (2000) 43: 194–202]