Since traditional intrusion detection methods perform poorly on unknown attacks and do not consider the impact of insufficient information on decision-making， an intrusion detection algorithm based on three three-way decisions and data augmentation called CGAN-3WD is proposed. The conditional generative confrontation nets are used to meet the data requirements of the three-way decisions. First， the three-way decisions theory is used to make decisions about network behavior， and it can catagorize network behavior into the positive domain， the negative domain or the boundary domain. Second， new samples are generated by the conditional generative adversarial nets， and the new samples can provide more information for the classifier to put the boundary domain into the positive or the negative domain. Third， the NSL-KDD dataset is used in the experiments. Experiments have proved that the CGAN-3WD model has indeed achieved good performance in intrusion detection， and it can effectively detect intrusions.