With the rapid development of our society, World Wide Web has turned to be an indispensible part of our daily life. Meanwhile, the network security is becoming more and more important. Intrusion Detection System (IDS), which serves to detect the abnormal activities in computers and internet, is often used to solve the network security problems. However, the IDS has to face and process the high dimensional data with high redundancy due to the increasing scale and dimension of the data, which causes the low efficiency of IDS. This paper proposes a new feature selection method for intrusion detection based on the Uniformed Conditional Dynamic Mutual Information (UCDMIFS), which can highly decrease the dimensionality and increase the detection accuracy. To examine our algorithm, the UCDMIFS algorithm is applied to the KDD Cup 99 data set and compared with other algorithms, such as support vector machine (SVM), to detect the intrusions. The experiments illustrate the efficiency of our algorithm.