The invention discloses a liquid level control method of a dual-loop water tank based on model prediction; the method comprises the steps: performing off-line recognition to the liquid level system of the water tank according to the openness of an input valve of the liquid level system of the dual-loop water tank and the historical data of the output liquid level, and establishing an overall non-linear mathematic model of the system depending on a liquid level non-linear auto-regression (ARX) model of the water tank; optimizing the parameters of the non-linear ARX model and confirming the order of the non-linear ARX model by utilizing an AIC (Akaike Information Capacity) criteria; designing a non-linear predication control algorithm by means of a standard quadratic performance index, obtained by the off-line recognition, based on the overall non-linear mathematic model depending on the liquid level non-linear ARX model of the water tank; acquiring a prediction controlled quantity v at each sampling moment by means of the global non-linear characteristic depending on the liquid level non-linear ARX model of the water tank; and realizing the liquid level control performance index by setting the parameters of a predication controller. With the combination of a system recognition technique and an automatic control technique, the method is a modeling and prediction control method with wide applicability for the liquid level system of the dual-loop water tank.