This paper shows how developments in the area of neural network combined with genetic algorithms can be used in the
handwritten digit recognition. In this work, two approaches to the design of a feed-forward neural network that model the
handwritten recognition system are discussed. The first approach focuses on constructing the network by using a trail-and error method the second approach is responsible for determining the appreciate parameters of the neural network and its
learning algorithm by the mean of genetic algorithms. Results show that using genetic algorithm for selecting the near
optimal parameters of the neural network, is improving classification performance on handwritten digits.