Knowledge distillation-based anomaly detection methods generate same outputs for unknown classes due to the symmetric form of the input and ignore the powerful semantic information of the output of the teacher network since it is only used as a "reference standard". Towards this end, this work proposes a novel Asymmetric Distillation Post-Segmentation (ADPS) method to effectively explore the asymmetric structure of the input and the discriminative features of the teacher network. Specifically, a simple yet effective asymmetric input approach is proposed to make different data flows through the teacher and student networks. The student network enables to have different inductive and expressive abilities, which can generate different outputs in anomalous regions. Besides, to further explore the semantic information of the teacher network and obtain effective discriminative boundaries, the Weight Mask Block (WMB) and the post-segmentation module are proposede. WMB leverages a weighted strategy by exploring teacher-student feature maps to highlight anomalous features. The post-segmentation module further learns the anomalous features and obtains valid discriminative boundaries. Experimental results on three benchmark datasets demonstrate that the proposed ADPS achieves state-of-the-art anomaly segmentation results.