When there are many fuzzy rules during Fuzzy-neural Networks (FNN) operation,the networks usually have slow learning speed and long running time.To solve this problem,a rough set theory was introduced to improve FNN model.Rough set data analysis method was used to obtain the reductive rules which were used as the fuzzy rules of the FNN.The input to the model was mapped into the output subspace by using these rules and the output of the system was approximated by improved BP algorithm training in this subspace.The results show that rules acquired by rough set data mining technology not only can get minimum rules but also are incomplete rules.Dimension of input and nerve cell numbers of network are decreased and learning speed is improved,which can meet time limitation of system.