In this paper, we propose a novel chaotic Hopfield neural network (CHNN), which introduces chaotic noise to each neuron of a discrete-time Hopfield neural network (HNN), and the noise is gradually reduced to zero. The proposed CHNN has richer and more complex dynamics than HNN, and the transient chaos enables the network to escape from local energy minima and to settle down at the global optimal solution. We have applied this method to solve a few traveling salesman problems, and simulations show that the proposed CHNN can converge to the global or near global optimal solutions more efficiently than the HNN.