The classical method of synthesis for control systems with state space model is reduced to finding linear or nonlinear feedback with respect to the system current state, which ensures asymptotic stability and some quality performances of closed system. In certain cases, the feedback is synthesized not on state variables, but on the outputs which are measured. If the system has a large dimension, we get very complicated control law, which is not always justified. In such cases, and under uncertainty may be used another approach. Using well-known methods of model order reduction it is possible appriory to construct approximating model of low order. And then on the base of truncated low-dimentional model of such systems synthesize control law. But in recent years have been developed an alternative method, which is based on the use of predictive models, which allow for sliding interval of some length to find the optimal or quasi-optimal control. This method is usually combined with the appropriate observer which also constructs using moving horizon estimation on sliding interval, but with other length. Besides for complex systems are used reduced models. It is planned to research this model predictive control and to demonstrate it’s possibilities on the examples.