Research in the area of Jawi character did not start until the middle 1980's. Some of the earlky work was done by Syed Yahya bin Mohd Salled at the Malaysia., University of Technology and significant progress in this field had been made. In this paper, a new approach fo reducing the number of classes to be recognized is explained. This research continue tp gain meomentum and was enhanced bu introducing Neural Netwerok as a tool for Jawi character recognition. Method of data presentation (raw pixel map entry) and choice of 3-layer neural network architecature will be examine. This thesis introduces a neural networks approach which is non model based, for the recognition of Jawi characters using computer packaged called mathematica. The approach was evaluated using isolated printed characters form a database established for this purpose. The results indicate that the technique offers an efficient and reliable classification rates.