Abstract: The heteroscedasticity is inevitable for the panel data modeling in economics. The two-stage estimation method is a better means to study the heteroscedasticity, in which the basis is to select only one independent variable for samples grouping, it can cause the information used is incomplete. In this paper, we propose to select several variables for grouping using variable selection method, then k-mean algorithm is used to cluster, so the samples classification can be achieved and the heteroscedasticity estimation can be obtained. The results of real example analysis show that the method presented in this paper has obvious advantages in effectiveness and feasibility.