In this paper, the chaotic lag synchronization for coupled time-delayed neural systems is investigated in detail firstly. We analyze the asymptotic stability for the error dynamical system based on Lyapunov method and linear matrix inequality (LMI) technique. A new sufficient condition for determining the lag synchronization between the coupling systems is derived. Above all, we skillfully shift our criterion which is expressed in the terms of LMI into the generalized eigenvalue minimization programming (GEVP) for the first time. The minimum of coupling strength is obtained successfully. A numerical experiment illustrates the effectiveness and advantage of our results