Stability of model predictive control could be achieved by adding a terminal weighting term in a performance index and the feasibility/stability region largely depends on the choice of the terminal weighting term and associated terminal control laws. For constrained linear systems, different from existing methods where the stability region is estimated by its terminal region, a new method to estimate feasibility/stability regions directly is proposed using a new representation of the behaviour of MPC. A design procedure is then developed to determine the terminal term such that the feasibbility/stability region of MPC algorithms is as large as possible. Examples show that the stability region is greatly enlarged.