In the chapter we discussed a control approach for torque control of gasoline engines. Because of several actuating variables and control requirements the process to be controlled is multivariable. The actuating variables are usually bounded and the effects on the engine torque are nonlinear. Hence direct use of the actuator variables for torque control generally produces plenty of problems. The two layer approach described in the chapter allows the application of standard control methods. The main idea of the control structure is to compensate or alleviate the nonlinearity behaviour by subordinate control circuits. All the physical actuating variables are substituted by setpoints of the subordinate systems. The torque controller so can be designed on base of linear models. Additionally only bounds of some control variables have to be considered. An appropriate standard control concept for the superordinate torque controller is the model predictive control principle. For the implementation we used a state space approach. The optimization problem is solved by the active set algorithm. For lower computation effort a solution with constant parameters was introduced. But this solution doesn't consider constraints and loss performance can occur. The three examples in section 4 show the capability of the control concept. The extension of the control structure is possible simply by completing the model and appropriate weighting parameters. In this way the controller should be able to handle more actuating variables or other requirements. The control quality depends on the quality of the model. Although the subordinate control circuits contribute to the linearization of the process behaviour the dynamic parameters may be dependent on engine speed or load. For that case a set of linear models could be useful. The control approach described in this chapter demonstrates that modern control approaches have considerable potential to improve the performance of embedded control systems. In addition to better performance also the variability of the systems and the ability to handle different control requirements could improve.