Consider a neural network model defined by the following state equations: where u = [u₁, u₂, ... ,u_n]^T is the neuron state vector, A = diag(a_i) is a positive diagonal matrix, &tgr;(t) is the transmission delay, W₀ = (w_ij⁰)_{n × n} and WM₁ = (w_ij¹)_{n × n} are the interconnection matrices representing the weight coefficients of the neurons, I = [I₁, I₂, ... ,I_n]^T is the constant external input vector, and g(u) = [g(u₁),g(u₂), ... ,g(u_n)]^T denotes the neuron activation. Under the assumption of bounded activation functions, this paper presents some new results ensuring the exponential stability of delayed neural networks with time varying delays. Shifting the equilibrium point u^* = [u₁^*,u₂^*, ... ,u_n^*] of the system (1) to the origin by the substitution x(.) = u(.) - u^* converts system (1) into the following form

By employing a Lyapunov-Krasovskii functional, and using a linear matrix inequalities (LMI) approach, a new exponential stability criterion is obtained. A sufficient condition for the global exponential stability of the equilibrium point for the delayed neural system defined by equation (2) is given. The paper compares the results obtained with well-known results in the literature.