We currently face a shortage of skilled network design engineers. Improving working efficiency and design quality of non-experts through design support systems based on machine learning or other techniques is therefore an important concern. We developed a network design support system that recommends similar configuration files when an engineer creates a new design. Our system uses a dictionary extracted from the network device manual to convert dependent areas for each setting in existing configuration files into a representation that shows design characteristics. It then clusters converted configuration files (unsupervised learning) and finds similar ones. We report on the concept, implementation and experimental results of the system.