The main purpose of this paper is to build the model about the earnings per shares of the public traded the listed companies on TSEC in 2007. The regression analysis is used. Traditionally, in regression models, predictors have been constructed using a parametric approach under the assumption that E(Y|X=x)=d(x, θ), where d has known functional form depending on x and a finite set of parameters θ=(θ1, θ2,…, θm). Then θ is estimated by the least squares method. In practical applications, however, the functional form of d is usually unknown. In such a situation, it is difficult to determine the regression function and we will use Classification And Regression Tress (CART) methods based on integrating the data and the model. We utilize selection procedures to select important predictor variables in the regression model based on data to predict Earnings per shares. Some criteria for selecting the important variables will also be discussed.