Both artificial neural networks and gene algorithm are a method that people imitates biological treatment pattern and gains the intelligence out of it in order to handle complicated problems. Based on uncertain network model structure and indeterminate initial weights and slow convergence speed for back propagation neural networks, this paper proposes that make use of gene algorithm to improve it in order to find the most suitable network connection weights and network structure, then form GA-BP model and apply it to currency recognition. The experiment indicates that GA-BP model shorten training time for Back Propagation Neural Networks and gain higher recognition speed and better recognition effect, thereby, it is preponderant for image recognition that make use of gene algorithm to improve back propagation neural networks.