The analysis of indirect immuno fluorescence (IIF) on human epithelial type 2 (HEp-2) cells is of paramount importance for the autoimmune diseases diagnosis. Essentially, accurate segmentation masks can generate rich boundary information, which is beneficial for improving the performance of the classification task For this reason, this paper proposes a novel segmentation guided HEp-2 cell classification method via generative adversarial networks (GANs), which employs the GANs as the segmentor to generate accurate masks for the subsequent classification task. Specifically, the proposed network architecture consists of three modules (i.e., the generator, discriminator and classifier). The first two modules constitute GANs model, which is trained to obtain better segmentation results via playing a min-max game. The segmentation masks and the corresponding original images are fed to the third module together to identify the category of the trained cell. Furthermore, the Xception and ResNet-50 model are used as the backbone of the segmentation and classification network, respectively. Besides, an improved classification loss function via Gaussian Mixture (GM) is proposed to optimize the classification network The proposed architecture can learn rich boundary information and well represent the class label of HEp2 cell images. Experimental results on the HEp-2 International Conference on Pattern Recognition (ICPR) 2016 task1 dataset demonstrate our proposed model achieves quite promising performance.