The paper introduces 3DSS-Mamba, a novel 3D-Spectral-Spatial Mamba framework for hyperspectral image (HSI) classification. It addresses the limitations of traditional CNNs and Transformers in capturing global spectral-spatial relationships with high computational efficiency. The framework integrates a Spectral-Spatial Token Generation (SSTG) module to convert HSI cubes into 3D spectral-spatial tokens and a 3D-Spectral-Spatial Selective Scanning (3DSS) mechanism to enable efficient global modeling. The 3DSS mechanism performs pixel-wise selective scanning along spectral and spatial dimensions, with five scanning routes to explore dimension prioritization. The 3DSS-Mamba block (3DMB) combines these components to extract global spectral-spatial semantic representations. Experimental results on three public HSI datasets (Pavia University, Indian Pines, Houston 2013) show that 3DSS-Mamba outperforms state-of-the-art methods in classification accuracy, achieving significant improvements in overall accuracy (OA), average accuracy (AA), and kappa coefficient (Kappa). The framework demonstrates superior performance in capturing long-range spectral-spatial dependencies with linear computational complexity, making it efficient for HSI classification tasks. The method is also robust to varying training sample proportions and outperforms other approaches in terms of computational efficiency and classification effectiveness. The research provides a feasible solution for HSI classification and highlights the potential of sequence scanning models in this domain.The paper introduces 3DSS-Mamba, a novel 3D-Spectral-Spatial Mamba framework for hyperspectral image (HSI) classification. It addresses the limitations of traditional CNNs and Transformers in capturing global spectral-spatial relationships with high computational efficiency. The framework integrates a Spectral-Spatial Token Generation (SSTG) module to convert HSI cubes into 3D spectral-spatial tokens and a 3D-Spectral-Spatial Selective Scanning (3DSS) mechanism to enable efficient global modeling. The 3DSS mechanism performs pixel-wise selective scanning along spectral and spatial dimensions, with five scanning routes to explore dimension prioritization. The 3DSS-Mamba block (3DMB) combines these components to extract global spectral-spatial semantic representations. Experimental results on three public HSI datasets (Pavia University, Indian Pines, Houston 2013) show that 3DSS-Mamba outperforms state-of-the-art methods in classification accuracy, achieving significant improvements in overall accuracy (OA), average accuracy (AA), and kappa coefficient (Kappa). The framework demonstrates superior performance in capturing long-range spectral-spatial dependencies with linear computational complexity, making it efficient for HSI classification tasks. The method is also robust to varying training sample proportions and outperforms other approaches in terms of computational efficiency and classification effectiveness. The research provides a feasible solution for HSI classification and highlights the potential of sequence scanning models in this domain.