In order to solve the difficult problem of handwriting postal codes recognition, an improved coarse grid feature extraction approach which simplifies network input of the neural network was introduced. BP neural network based on LM algorithm for network studying was adopted. LM algorithm is an improved Gauss-Newton algorithm. The improved algorithm further enhances precision, stability and studying speed of the network studying through the simplification of the network input. The simulation results show that the algorithm is effective on the handwriting postal codes recognition.