AbstractThis paper presents an artificial intelligence approach to solving the network design problem. The major issue under consideration is the various topological aspects of a design, especially in the case of large computer networks. The combinatorial explosion can only be dealt with, hopefully, by means of heuristics, which drastically reduce the search space of candidate topologies. The proposed approach is essentially a general framework for integrating conventional approximate methods and knowledge-based systems. An inference engine is then a means suited to extending the realm of local transformations used in conventional heuristics. Sophisticated goal-directed searches can be achieved by using knowledge bases, which are large sets of detailed design rules.