A network (CC-Net) based on complementary consistency training is proposed for semi-supervised left atrial image segmentation in this paper. From the perspective of complementary information, CC-Net efficiently utilizes unlabeled data and resolves the problem that semi-supervised segmentation algorithms currently in use have limited capacity to extract information from unlabeled data. A primary model and two complementary auxiliary models are part of the complementary symmetric structure of the CC-Net. The inter-model perturbation is formed between the main model and the auxiliary model to form complementary consistency training. The complementary information between the two auxiliary models helps the main model to focus on the fuzzy region effectively. Additionally, forcing consistency between the main model and the auxiliary models makes it easier to obtain decision boundaries with low uncertainty. CC-Net was validated in the benchmark dataset of the 2018 Atrial Segmentation Challenge. The Dice reached of 89.82% with 10% labeled data training and 91.27% with 20% labeled data training. By comparing with current state-of-the-art algorithms, CC-Net has the best segmentation performance and robustness. Our code is publicly available at https://github.com/Cuthbert-Huang/CC-Net.