Accurate tooth CBCT segmentation is a crucial step in providing complete dental structure information for computer-aided diagnosis. Despite the emergence of many fully supervised deep learning tooth segmentation methods in recent years, annotating all teeth is a time-consuming and laborious task due to the large number of teeth and the similarity of tooth roots to surrounding bone density. Therefore, the application of semi-supervised learning in tooth segmentation has received increasing attention from researchers. In this study, we propose a semi-supervised tooth segmentation method combining the entropy-guided mean-teacher (EG-MT) and the weakly mutual consistency network (WMC-Net). Diverging from MC-Net \(+\) , we replace the last up-sampling layer with two up-sampling layers using different up-sampling methods on the basis of V-Net to provide consistency information and enhance local information attention through the CBAM module. The EG-MT strategy is designed to effectively guide the network in learning pixels that are difficult to recognize, such as boundaries. In addition, we perform post-processing operations such as erosion and dilation on the segmentation results to improve accuracy. On the validation set, we achieved an average DSC of 88.39%, an average IoU of 89.32%, and an average HD95 distance of 0.0934. On the test set of the MICCAI STS 2023 Challenge, we achieved an average DSC of 77.49%, an average IoU of 81.61%, and an average HD95 distance of 0.1580. Our code is available at https://github.com/59-lmq/STS2023-WMCNet .