Combining boosting and Support Vector Machine (SVM) is proved to be beneficial, but it is too complex to be feasible. This paper introduces an efficient way to boost SVM. It embraces the idea of active learning to dynamically select “important” samples into training sample set for constructing base classifiers. This method maintains a small training sample set with settled size in order to control the complexity of each base classifier. Other than construct each base SVM classifier directly, it uses the training samples only for finding support vectors. This way to combine boosting and SVM is proved to be accurate and efficient by experimental results.