In this thesis, we present an extended account of the robust statistical inference for the linear models based on the article by Salibian- Barrera (2005). There are several proposals of robust estimates, but between them, MM estimates have more suitable robust properties. MM estimates have simultaneously high efficiency and high breakdown point. We focus on the problem of estimating p-values for robust scores-type tests. To estimate the p-value of robust tests under general conditions, we can use the bootstrap. Since the robust tests are based on robust regression estimates thus, to bootstrap these robust tests we need to bootstrap the corresponding regression estimates. In this thesis, we introduce a new method to estimate the distribution of robust regression estimates. The idea based on bootstraping a reweighted representation of the estimates. This method which we call Robust Bootstrap (RB) is computationally simple, because for each bootstrap sample we only have to solve a linear system of equations. The applied weights decrease the absolute value of the residuals. In this way, the outlying observations receive small weights. This thesis is organized as follows. In chapter one, we have a history and a brief introduction. Chapter two deals with discussing the necessary definitions and concepts. Then, in chapter three, we present the definition of the robust regression, introduce different kinds of robust estimates and compare them with each other. In chapter four, we introduce different kinds of robust tests and details of Robust score test based on MM estimator. In chapter five, we study the ltr"