In this paper, the main idea is to compute the robust regression model, derived by experimentation, in order to achieve a model with minimum effects of outliers and fixed variation among different experimental runs. Both outliers and none quality of residual variation can affect the response surface parameter estimation. The common way to estimate the regression model coefficients is the ordinary least squares method. The weakness of this method is its sensitivity to outliers and specific residual behavior, so we pursue the modified robust method to solve this problem. Many papers have proposed different robust methods to decrease the effect of outliers, but trends in residual behaviors pose another important issue that should be taken into account. The trends in residuals can cause faulty estimations and thus faulty future decisions and outcomes, so in this paper, an iterative weighting method is used to modify both the outliers and the residuals that follow abnormal trends in variation, like descending or ascending trends, so they will have less effect on the coefficient estimation. Finally, a numerical example illustrates the proposed approach.