A histopathological image is a microscopic image applied to examine cellular and tissue structures and identify any abnormalities or disease processes. Histopathological image segmentation is a prerequisite step for analyzing histopathological images that can divide an image into meaningful regions or objects to accurately classify and analyze tissue structures, cellular regions, or particular histological entities. However, the existing deep learning based pathological image segmentation methods require huge annotation efforts from the pathologists, which is labor-intensive and time-consuming. In this scenario, it has become a hotspot to leverage abundantly available unlabeled data to help learn segmentation models given limited labeled data. In this paper, we propose a global–local consistent semi-supervised segmentation (GLCS) model that enforce the consistency of the segmentation results with weak and strong perturbations on unlabeled data. In GLCS, we firstly generate different weak perturbations for each unlabeled sample, and then add a regularization term to ensure the segmentation consistency among different weak perturbations. Next, different from the existing studies applying the regression methods to match the segmentation results among different perturbations, our methods are based on the generative adversarial learning that can keep the global structure consistency among unlabeled data with different strength of perturbations. Finally, we also add a patch-correlation based regularization term to preserve the local structure similarity among different perturbations images. We validate our GLCS on three datasets, i.e. Glas, Crag and MoNuSeg. The experimental results show that our method can achieve to the dice ratio of 90.35, 82.61 and 81.60 with 1:1 proportion of labeled data, which are significantly superior to the state-of-the-art semi-supervised histopathological image segmentation methods. Our code is public available at https://github.com/ISBELLAG/GLCS.