Model-based control systems are better fitted to glycaemic control in intensive care than ad-hoc protocols, but depend on predictive accuracy and facilitation of clinical routines. A general method to customize and visualize modelbased blood glucose predictions is presented. Customization is based on admission type and diabetic status of patients. Blood glucose concentrations of 14 critically ill patients from two intensive care units were retrospectively predicted. Relative prediction errors were found to be highest for diabetic I and II patients, and lowest for non-diabetic trauma and head-injured patients. Standard deviations of mean relative prediction errors are proposed to be used for display of accuracy of model-based blood glucose predictions in prospectively controlled patients. The method provides for an optimized timing of blood sampling to facilitate tight glucose management in the ICU.