On the basis of the connective mode and information transmit mechanism of cerebral cortex, a novel lateral inhibition neural network model with span connection(S-LINN) is proposed. Combining the liminar organization of cerebral cortex, and fully considering the lateral inhibition connection between interneurons and the pyramidal neurons, the proposed S-LINN transforms information to other neurons in different layers, which is used to enhance response contrast and advance the network representation, respectively. The effectiveness and superiority of the proposed network is compared with other popular approaches on two benchmark problems in the areas of real-world regression and classification problems. Simulation results show that the proposed S-LINN can achieve the higher accuracy of approximation and generalization with the comparably compact network structure.