This paper analyses the weakness of setting a single, specified threshold for the size of neighborhood, and then puts forward a new neighborhood setting method based on the standard deviation of feature data. The paper introduces ant colong opbimization (ACO) into feature selection and proposes an approved feature selection algorithm based on NRS and ACO, in which the feature importance is taken as the heuristic information. In order to evaluate the performance of the proposed algorithm, four datasets from UCI are used and the experimental results show that the proposed algorithm has a better performance in classification accuracy of reduct and feature number in reduct.