We consider a hybrid selection and testing design for comparing the means of several experimental normal populations among themselves and with the mean of a control normal population. It is assumed that the variances of the experimental and the control normal populations are unknown and unequal. A Stein-type two-sample selection is used at both selection stage and the testing stage to solve the heteroscedastic problems caused by unknown variances. The hybrid two-stage design allows for (1) dropping the poorly performing treatments early on the basis of interim analysis results and (2) allowing for early termination if none of the experimental treatments seems promising. Numerical computations are given to show the advantage of the proposed procedure over a pure selection procedure.