This paper presents an overview of neural network (connectionist) systems and their potential contribution to computer-aided design. The authors discuss the appeal of neural networks and some of the problems. The major contribution to design is in the representation and manipulation of schemas. A neural network system can be'taught'various examples (such as room descriptions). The system then apparently recognizes schemas (room types) and can produce novel but sensible combinations of descriptions constituting new types. A simple handworked example is presented, and the learning and reasoning mechanism is explained