To elucidate abnormalities in the feedback relationships between plasma glucose and plasma insulin levels in diabetic patients, we have introduced the vector autoregressive modeling method as a new for tool feedback analysis. This technique was applied to plasma glucose and insulin level data from a series of 977 frequently-sampled oral glucose tolerance tests (FS-OGTT). Neither special instruments nor medications were used in FS-OGTT. We were able to predict the degree of the plasma glucose response occurring after an impulse-like increase in plasma insulin at 1 mU/mL, as well as the plasma insulin response triggered by an impulse-like increase in plasma glucose at 1 mg/dL, in the form of "impulse response curves". The predicted impulse response curve of glucose to insulin gradually changed from negative to positive with incremental changes in the fasting plasma glucose level, reflecting increased insulin resistance. Furthermore, the response of insulin to glucose decreased in a stepwise fashion with the incremental changes in the fasting plasma glucose level. Our findings confirm the usefulness of impulse response curves as clinical indicators. In addition, analytical data point to a possible contribution of excessive hepatic glucose production to the pathogenesis of the insulin resistance in non-insulin-dependent diabetes mellitus.