Artificial neural networks(ANN)and genetic algorithm(GA)are both theories for intelligent information processing,but there are limitations in these algorithms. Combining genetic algorithm with BP neural networks,an improved algorithm for image compression using a mixed BP neural network(BP-GA)is proposed. It looks on the network weights as a unitary chromosome that is operated by the three genetic operators including selection,crossover and mutation as the normal genetic algorithm. The experimental results show that the mixed algorithm can avoid premature saturation and have higher compression rate and better effect in image compression.