Abstract: The separation of structure and algorithm in NN simulating model and how to reuse the code of NN through language description are introduced. Through comparing meny kinds of neural networks, the rules for NN to follow are concluded the feasibility of separating the structure and the algorithm is demonstrated, and the ideal boundary of the separation and the method of appropriate description and combination of these two parts are given. Users need only give the description of the new structure designed, then select an appropriate one in the algorithm library, and a neural network program will be generated automatically. Up to now, such a method of separating structure and algorithm is considered having advantage and significance.