In order to improve the recognition rate of similar samples by NFS (neural-fuzzy system), bagging algorithm was proposed to improve the recognition rate. As the bagging algorithm needs a simple basic classifier, the basic classifier needs to be modified. In this paper, a new NFS was obtained when the traditional NFS input layer was removed. Then, with the combination of the bagging algorithm and a new NFS, a new model was set up. The experimental results showed that the recognition rate of the new model was not only 1.67% higher than that of a single NFS, but also the same as that of the decision tree, softmax and xgboost on Iris dataset. Based on the sensitivity and specificity analysis of the new model, the linear data can get better result of classification than the non-linear data. This new model as obtained by combining bagging with the new NFS features rapid prototyping, strong generalization ability and high recognition rate.