Preliminary results of an in-house predictive control algorithm applied to blood glucose level control in type I diabetic patients are presented in this paper. The manipulated variable considered in this application is the insulin infusion rate to a drug delivery pump; the controlled variable is the blood glucose level. The role of the disturbance filter in model-based predictive control is indicated and its possibility to improve control performance is described. Controller performance was assessed in terms of its ability to reject the effect of 60g glucose as meal disturbance, on an averaged nonlinear patient model. Unconstrained control had reasonably good and stable results within imposed constraints. Robustness was tested over a group of 7 patients with model parameters varying considerably from the averaged model. The performance of the controller is critically assessed with some further ideas for improving the control robustness and advantages of applying nonlinear predictive control are given