Based on the research of the functions of ANN-based cold extrusion process design system, genetic algorithm (GA) is proposed to optimize the topology and parameters of artificial neural networks (ANN), in order to improve the running efficiency of the networks. The binary encoding approach is implemented to represent the GA chromosome. The code string or the chromosome was divided into three parts: the first part is the binary code of the cold extruded part; the second part is the binary code of the topology and parameters of ANN; the last is the binary code of the semi-cold-extruded-part or the billet. The 1/F(X) function is selected as the fitness function in GA, where, X represents the binary code of the cold extruded part, F(X) represents the error between the real outputs of ANN and the desired results; the biased roulette wheel selection method is used for selecting operation in this paper; two-point crossover and one-point mutation are selected for these two types of genetic operations. Finally, the typical cold extruded part is used for verification as an example by using the optimized ANN, the result shows that ANN optimized by GA has efficiency and validity in the cold extrusion process design system.