The control of linear time invariant systems including unknown parameters has been one of the most important and interesting fields of systems theory. Today, it has been generally accepted that when the parametric error is small,one identification model can afford to deal with these kinds of systems providing desired and robust stability and performance. However, when the parametric error is high, this satisfaction could not be obtained because of the existence of high gain oscillatory transient response in adaptive systems. Till now, several researches have been done to improve the mentioned problem. One of the most successful methods involving multiple models introduced in 1990. During this period, both switching between multiple fixed models, and switching and tuning between fixed and adaptive models have been proposed. It has been agreed that the switching and tuning presents a good performance if there is no limitation on the number of used models. Generally, in these methods the number of necessary models to assure that at least one of the fixed models is close enough to the plant in parametric space is high and grows exponentially as the dimension of unknown parameters vector increased. In addition, various models do not share in any way while making decision about the location of parameters vector of the plant. The information and performance indices of various models are just used to determine a model which is the closest to the plant. So, the information obtained from a lot of models could not be used cooperatively and effectively. The conventional adaptive methods could not afford to present good performance in noisy conditions or when the parameters have fast variations. Recently, a new kind of multiple models has been proposed which has an important difference with the previous ones. The approach needs only n+1 models which is so smaller than (c is an integer number) when n is large. While each of the n+1models produces an estimation of the plant parameter vector, the final estimation generated by the new approach is dependent on the collective outputs of all the used models. This can be viewed as a time varying convex combination of the estimates. In comparison to the current multiple models methods, the new approach has a faster convergence. In this thesis, the new approach in multiple models has been developed in two fields. First, this approach has been applied to adaptive control of a linear time invariant model when the coefficients vector of characteristic polynomial of the plant is a linear affine function of an unknown vector. Since the affine linear transformation of polytops preserves the convex hull property, it is possible to estimate the unknown parameters using convex combination of the multiple models. Second, the approach has been applied to adaptive control of a ltr" Key words: Multiple models, Convex combination, Polytopic, Nonlinear systems, Adaptive control