The purpose of this paper is to introduce a multimodel approach to control uncertain systems. The basic problem is to minimize the number of models in the dynamic library in order to obtain the desired performance. A proposed solution is based on the Kharitonov principle. Each generic model and its corresponding controller have an unchanged structure. The multimodel control is obtained by fusion of the elementary controllers. A new strategy for multimodel control is proposed, which uses observers to improve the robustness of multimodel control in discrete systems. The principal idea of this control is to reinitialize the state of each model by the corresponding state computed from the process when the error between them exceeds a certain value. A good choice of observer, as well as a fusion applied to the coefficients of each controller, allows the multimodel control strategy to reach the desired objective for the multimodel control of the system. The desired performance is reached by the use of this new strategy of multimodel control. Thus, the significant aspect of this study is the introduction of observers into multimodel control as a technique to improve the validity of the generic models. The multimodel control obtained by this new strategy of control is presented. Computer simulation results show that a robust multimodel control design, based on an appropriate selection of models, plus reinitialization steps based on observers, is feasible and provides good performances.