This paper proposes a hybrid convolution and attention network (HCANet) for hyperspectral image (HSI) denoising. HSI denoising is crucial for effective analysis and interpretation of hyperspectral data, but modeling both global and local features remains underexplored. HCANet combines the strengths of convolutional neural networks (CNNs) and Transformers to enhance HSI denoising. A convolution and attention fusion module (CAFM) is designed to capture long-range dependencies and neighborhood spectral correlations, while a multi-scale feed-forward network (MSFN) improves multi-scale information aggregation. Experimental results on mainstream HSI datasets demonstrate the effectiveness of HCANet in removing various types of complex noise. The proposed model outperforms state-of-the-art methods in terms of quantitative metrics and visual quality. HCANet achieves optimal denoising performance with a relatively moderate number of parameters and computational complexity. The model is trained on the ICVL dataset and evaluated on the Pavia dataset under different noise settings, including Gaussian, impulse, and deadline noise. Ablation studies show that each component of HCANet is essential for effective denoising. The model successfully restores original image features while preserving local details. The code is available at https://github.com/summitgao/HCANet.This paper proposes a hybrid convolution and attention network (HCANet) for hyperspectral image (HSI) denoising. HSI denoising is crucial for effective analysis and interpretation of hyperspectral data, but modeling both global and local features remains underexplored. HCANet combines the strengths of convolutional neural networks (CNNs) and Transformers to enhance HSI denoising. A convolution and attention fusion module (CAFM) is designed to capture long-range dependencies and neighborhood spectral correlations, while a multi-scale feed-forward network (MSFN) improves multi-scale information aggregation. Experimental results on mainstream HSI datasets demonstrate the effectiveness of HCANet in removing various types of complex noise. The proposed model outperforms state-of-the-art methods in terms of quantitative metrics and visual quality. HCANet achieves optimal denoising performance with a relatively moderate number of parameters and computational complexity. The model is trained on the ICVL dataset and evaluated on the Pavia dataset under different noise settings, including Gaussian, impulse, and deadline noise. Ablation studies show that each component of HCANet is essential for effective denoising. The model successfully restores original image features while preserving local details. The code is available at https://github.com/summitgao/HCANet.