In the problem of scheduling data flow computers, the optimization of scheduling needs much cost. The cost often increases in non-polynomial order as the number of the processing elements and tasks increases. Many studies have been tried to decrease the order. We have used Hopfield-type neural network model for this problem. We have achieved the reduction of calculation cost in proportion as the number of tasks and the number of processing elements. Our algorithm schedules flow graphs onto the data flow architectures. Our algorithm optimizes communication delays and parallel processing delays only on the critical path of the graph. So the result of our scheduling is more superior to therecent other studies which employs neural network models. Our method is so good at being processed in parallel for it employs neural networ model.