Model-based tight glycaemic control is primarily a function of model quality. A new, more physiologically relevant control model is presented and validated over a 270 patient data set. Prediction and fitting errors lie within measurement error, indicating a suitability for clinical use. A stochastic model is developed and shows the same trends expected in the dynamic, critically ill patient who gets well slowly, can decline quite rapidly in some (more rare) cases. The combination of these models is suitable for use in acute and critical care settings This deterministic + stochastic model combination has already been applied successfully in 24-hour NICU trials. Models & Methods