With respect to multiple attribute group decision making (MAGDM) problem where the evaluation information is in the form of trapezoidal twodimension linguistic numbers (T-2DLNs), a new similarity measure for T-2DLNs is proposed, and its properties are discussed as well. The two optimization models are constructed to solve experts’ weight and attributes’ weight in MAGDM on the basis of maximizing similarity measure, and then a new MAGDM method via similarity measure for T-2DLNs is presented. Finally, a numerical example indicates that the proposed method is effective and feasible.