The most effective control algorithm for regulation of glucose levels in persons with diabetes is determined using control-relevant nonlinearity analysis of a diabetic system model. Theoretical control-relevant nonlinearity analysis is performed using the Optimal Control Structure and a norm-based nonlinearity measure. These results are correlated with results of controller performance assessment trials, based on optimizing the rejection of glucose disturbances. Performance is quantified using a standard quadratic performance objective as well as an asymmetric performance objective in which negative glucose deviations are penalized more highly than positive deviations because of the greater health concerns associated with negative deviations. The control-relevant nonlinearity assessment indicates that the best controller design is a linear algorithm except when the desired performance is strongly asymmetric. For standard meal disturbances, the system is found to be well regulated using proportional-derivative control or standard linear model predictive control with no significant benefit observed in using nonlinear model-based control. (C) 2005 American Institute of Chemical Engineers.