As we know, feature selection can improve the performance of machine learning algorithms for intrusion detection. This paper proposes a hybrid feature selection method, which ranks features according to two factors: relevancy and redundancy, and then adopts the forward search strategy to select the optimal feature subset from the ranked features. Experiments on the KD-DCup' 99 dataset showed that our proposed feature selection method could get better performance on the accuracy rate and false positive rate in intrusion detection compared with other feature selection approaches.