We have tested the hypothesis that hepatic glucose output (R(a)) during exercise in humans is subject to feedback control by circulating glucose within a control range that is determined by the circulating insulin concentration. Three exercise protocols based on 60-min cycle ergometer exercise at 55% maximal O2 consumption were used: 1) control, 2) insulin infusion with a euglycemic clamp, and 3) insulin infusion with a fixed-rate glucose infusion. R(a) was measured using a constant infusion of [3H]glucose. During the glucose clamp there was no R(a) response to exercise. There were significant inverse relationships between R(a) and plasma glucose during control exercise (r = -0.73, P < 0.001) and exercise with fixed-rate glucose and insulin infusion (r = -0.96, P < 0.001). During the fixed-rate glucose and insulin infusion, plasma glucose fell from the commencement of exercise but stabilized at a lower level. These results are interpreted in terms of a simple difference controller where R(a) is proportional to the deviation of plasma glucose from a defined set point. Insulin affects R(a) and regulates the steady-state glucose level by altering the sensitivity of this control system.