The optical recognition system is based on the optical characteristic extractor. In this report, a kind of new theory of the characteristic recognition system with artificial neural network is introduced. The optical compound eye system, lateral inhibition network and back propagation network (BP) are adopted to form a parallel neural network, recognition system. The field of view is divided into mosaic pixels by the plane compound eye lens, which is convenient to use single photoelectric detector. The information received by the detector is extracted characteristic through the lateral inhibition network. It is a parallel neural network made up of resistor network and it has the advantage of high speed, simple structure, etc. BP network is used for pattern recognition. Its weights are anew distributed during network learning processing. Once the studied object is detected again, the system will quickly response its pattern. In this paper, several experimental data of simple patters are given, and the precessions of the network recognition are analyzed. Finally, it is pointed out that the characteristic recognition system is feasible in applying to industrial detection and Chinese character recognition, etc.