Feature extraction is one of the most critical steps of the classification system. This paper makes a deep research on the feature extraction of classification of voltage sag sources. Firstly, three feature extraction schemes are proposed based on Hilbert-Huang Transform (HHT) and CAIM feature discretization method, and then the three feature extraction schemes are tested on three classifiers, which are Decision Tree (DT), Probability Neural Network (PNN) and Support Vector Machine (SVM). Simulation results show that effective classification feature vector can be extracted by means of HHT-based feature extraction method, and discretization of feature vector could compress sample set effectively without reducing classification accuracy. Meanwhile, such process also enhances the robustness of the classification algorithm. All of them are important to realize accurate and fast recognition of voltage sag sources. This work is supported by National Natural Science Foundation of China (No. 51477105).