The Lineal Multiple Regression (RLM) is a method of analysis of data very popular among engineers and scientifics. This investigation studies the problem of the selection of variables in RLM but, with applications in data that present a marked multicollinearity. The theoretical bases of the method of selection of variables proposed by Mansfield, Webster and Gunst (MWG, 1977) are analyzed in that which of utility when the multicollinearity is problem. The method is implemented in a script of the statistical package R and it acting is compared with the denominated classic procedures of selection of variables.