When evaluating decoy effectiveness by means of BP neural network, training sometimes failed because of local extremum problem. The genetic algorithm neural network model for evaluating camouflage effectiveness of decoy is created for this purpose. Two steps of evaluating by this method is necessary and a series of index is put forward. After initializing weights and executing genetic operation, we finally retrain the network to get the results which show that the method has fast convergence and the model reliable, effective and objective. This paper is meaningful to camouflage theory and application.