Communication emitter identification has been recognized as an indispensable task for electronic intelligence system. Recognition of communication emitter signals is presupposition of jamming or striking enemies' equipment effectively in electronic warfare. This paper mainly studies the classification and identification of emitter signals. Firstly, the eigenvectors are reduced by principal component analysis (PCA) to realize the feature extraction of signal samples. On this basis, a flexible clustering model, integrating K-means clustering and hierarchical clustering is proposed to distinguish different types of signal samples, which overcomes the disadvantage of inaccuracy due to single algorithm in signal classification. Then a hybrid classification model with k-nearest neighbors, random forest and neural network is constructed to identify individual signals, and the accuracy is higher than 97%. Finally, artificial signals are generated by MATLAB to test our clustering model. The experimental results show that the clustering model is efficient and reliable, and the average accuracy is up to 93.4%.