This paper describes algorithm to extract license plates in vehicles. Neural networks have been trained to perform complex functions in various fields of application including pattern recognition, identification, classfication, speech, vision and control systems. Today, neural networks can be trained to solve problems that are difficult for conventional computers or humanbeing. The recognition methods using neural networks have two ones. In this paper, First, we input the normalized self-image and learned the real image. Second, we extracted charastic vector and learned it. But, the former is ideal method that is required many units according to resolution. Also, in increasing interconnection number, it takes many time to recognize an image. The latter has a decreasing problem of recognition and merits to reduce units and interconnection number in a few input. Simulation shows that the proposed algorithm having character recognition system using multi-layer neural networks improves the recognition rate in various learning data.