This paper describes the development of a computer system that determines process names, machine tools and so on in production system design. We have proposed multiobjective evaluation reasoning including neural networks learning. The reasoning acquires utility functions by using knowledge database for machine tools. Machine tool costs that are not clarified and are not also beforehand described in the knowledge database are presumed by the neural network and used in multiobjective evaluation reasoning. The networks have three—layered neuron construction. Because of the networks, the system can quickly presume the rough costs for machine tools.