This chapter presents a comprehensive overview of graph-based methods for hyperspectral image analysis, focusing on their ability to model complex spectral and spatial relationships in non-Euclidean domains. It first introduces the foundations of graph representation and construction strategies tailored to hyperspectral data. The chapter then discusses graph convolutional networks and their variants, including methods that improve graph structure, information propagation, and learning objectives. Following this, autoencoder-based models such as graph autoencoders and variational graph autoencoders are explored for unsupervised representation learning. Finally, the chapter examines attention-based networks and graph transformers, which enhance both local and global feature modeling through dynamic weighting and global context integration. Major challenges are analyzed in detail, and the chapter concludes with an outline of future research directions. Together, these developments highlight the growing maturity and versatility of graph-based approaches in advancing intelligent hyperspectral image analysis.