Abstract: In this paper, a design method for network fault diagnosis systems is put forward by proposing RSNN algorithm, which tightly combines neural network and rough sets Reduced information table can be obtained, which implies that the number of evaluation criteria is reduced with no information loss through rough set approach And then, this reduced information is used to develop classification rules and train neural network to infer appropriate parameters The rules developed by RS neural network analysis show the best prediction accuracy if a case does match any of the rules It’s capable of overcoming several shortcomings in existing diagnosis systems, such as a dilemma between stability and redundancy The experiment system implemented by this method shows a good diagnostic ability