: Diabetes is a disorder of body metabolism of blood glucose by insulin. In type 1 diabetes, pancreatic ?-cells that are responsible for the secretion of insulin in the body are destroyed by the immune system, so patient has to inject insulin. In traditional way to control type 1 diabetes, doctor by taking a regular day profile of blood glucose determines the dose of insulin that is injected intravenously several times a day. The disadvantage of this method is that it limits the patient to a particular life style of food and physical activity. Because by changing these factors, prescribed insulin dose would not be appropriate and hyperglycemia or hypoglycemia may be seen. In addition, some factors that influence blood glucose profile, such as stress, excitement and etc. are not under our control and may cause undesired operation of this way. The hypoglycemia short time adverse effects include blurred vision, loss of consciousness, coma and even death. The hyperglycemia in the long term lead to vascular disease, kidney and eye malfunction. According to the ineffectiveness of this method and mentioned problems, the importance of closed-loop and adaptive control of glucose _ insulin system that is the subject of this thesis is obvious. Closed loop control presented in this thesis consists of three parts: patient, identification and control algorithm. The challenge involved in this work is intense effect of food disturbance on system and also delay of subcutaneous injection and delay of continuous blood glucose sensor. In this study, from physiological models provided for the person with diabetes, Hovorka model is used as a virtual patient. The on line identification system is used to identify the effects of foreign factors that influence glucose _ insulin dynamic, that we cannot control them in traditional way, so the system is adaptive. In fact, in this step recursive least squares method is used to make a mathematical model (ARX and Hammerstein) by the data received from the virtual patient and model is updated every ten minutes. Then, in controller block by using model predictive control, we try to set blood glucose within the desired band with presence of delay. In this step by applying GPC and NMPC controllers at identified models, it is seen that NMPC has better performance but it droppes into hypoglycemia range for a short time. The conventional method to improve the performance of the system against food disturbance consist of adding feedforward control that need to manually enter the information of food that is eaten. Finally, unlike conventional method, we add glucose as a new control input that is equivalent of injecting glucose serum, and use multi-variable control system to improve the performance of the controller. In this situation, the hypoglycemia after applying food disturbance is removed but two inputs NMPC has better performance than two inputs GPC, so it is introduced as ideal controller. Key Words- Diabetes, Control of blood glucose concentration level, Adaptive control, System identification, Model Predictive control