Minimal model analysis of glucose and insulin concentrations in the intravenous glucose tolerance test (IVGTT) has been widely used to obtain a measure of insulin sensitivity in humans. Issues of model validity and IVGTT protocol have been explored extensively. Less attention has been paid, however, to the computer programming protocol for estimating the model parameters (programming implementation). Minimal model analysis of data from an IVGTT protocol involving a high glucose dose (0.5 g/kg) and a reduced sample schedule, employed in healthy pre- or post-menopausal women, healthy men or men with coronary heart disease or chronic heart failure (20 in each group), was undertaken according to 12 different programming implementations using a commercially available model-equation-solving program. The ability of the program to arrive at an acceptable solution to the model equations gave a success rate of between 39% and 96%, depending on the implementation. Variation in basal glucose assignment significantly affected the magnitude of estimates of insulin sensitivity. The maximum modelling success rate was achieved by introduction of an imputed glucose measurement at 360 min from the glucose injection, taking the basal glucose level as the fasting glucose concentration, and overweighting the initial glucose measurement after a delay for mixing. Use of this implementation to analyse data from a study comparing insulin sensitivities obtained using the minimal model and a euglycaemic clamp reference gave a correlation of 0.80 (P<0.001) between the two methods. Straightforward variations in programming implementation, involving appropriate assignment of the basal glucose concentration and use of an imputed glucose measurement signifying re-establishment of basal glucose levels following the IVGTT, can considerably improve modelling success rate.