The ancient way of sign language is most natural forms of communication. The recognition of sign is place a key role in research field. The development and improvement on this kind of work need more and more new techniques to analyze the accurate results. Many people don't know it and interpreters are hard to come by, we developed a real-time technique for finger spelling-based American Sign Language using neural networks. In our technique, the hand is first sent through a filter, and then it is passed through a classifier, which analyses the class of hand movements. For each alphabet the proposed model has a 96 percent accuracy rate. This model mainly implemented for Dumb and Deaf people for communication.