Abstract: ‍ ‍Electroencephalography （EEG） is a powerful technology used to reflect the electrophysiological activity of the cerebral cortex. It is frequently employed in the field of biomedicine for a variety of applications. However， during the acquisition process， EEG signals are often contaminated with various artifacts， such as muscle activity， eye movements， and heart activity. These artifacts can significantly degrade the quality of the EEG signals， which can adversely affect the subsequent analysis. As a result， artifact removal is an essential step in EEG signal analysis. As of now， deep learning-based methods have demonstrated better denoising performance than traditional methods in several benchmark datasets. However， existing deep learning-based methods still have the following limitation. Specifically， the existing network structures are not fully considered temporal characteristics of the signals. Different artifacts have distinct morphological characteristics， so it is challenging to obtain ideal denoising performance on various artifacts by relying solely on either local or global temporal dependencies. To address the above issue， this paper designed a novel EEG denoising network， named CTNet. CTNet adopted the CNN-Transformer structure， which combined the advantages of convolutional neural networks （CNNs） and Transformers to extract potential discriminative features. The CNN unit and the transformer unit are responsible for extracting local and global temporal features， respectively. The network obtained better artifact suppression ability by combining local and global temporal features. To evaluate the performance of CTNet in EEG denoising tasks， we evaluated the performance of CTNet on four different artifact removal tasks. The experimental results demonstrated that CTNet had strong noise suppression ability under various noise conditions. Furthermore， the proposed network outperformed existing methods and obtained the lowest root mean square error （RRMSE）， maximum correlation coefficient （CC）， and maximum signal-to-noise ratio （SNR） on various artifacts. For example， in the task of removing electromyography artifacts， CTNet achieved an 8.67% reduction in RRMSE， a 1.41% improvement in CC， and a 7.06% improvement in SNR compared to NovelCNN. In conclusion， the experimental results verified the superiority of CTNet in EEG denoising tasks， making it a promising approach for improving the quality and reliability of EEG analysis.