Abstract: In this study we present an off line system for the recognition of the isolated handwritten Arabic characters. The study is based on the analysis and the evaluation of multi-layers perceptron performances, trained with the gradient back propagation algorithm. It is hoped that the results of the evaluation contribute to the conception of operational systems. The used parameters to form the input vector of the neural network are extracted on the binary images of the characters by the following methods: the centerd moments of the projections sequences, distribution parameters, the Barr features and Coding according the directions of Freeman.