Purpose/Significance The traditional methods for segmenting pleural effusion heavily rely on prior knowledge, are cumbersome in process, time-consuming, and often exhibit poor performance. There is a need to enhance efficiency and accuracy in addressing these issues. Method/Process Based on the characteristics of pleural effusion in chest CT images, the paper proposes a pleural effusion segmentation model called ARU-Net, which is based on the residual attention mechanism. The ARU-Net model utilizes the U-Net architecture as its backbone network. It introduces residual attention units in both the encoding and decoding stages to effectively capture contextual information from the images, thereby improving the utilization of features. Result/Conclusion The DICE similarity coefficient on the test set reaches 88.76% for ARU-Net, and shows significant advantages in segmentation integrity and accuracy compared to U-Net and ResU-Net, which can meet clinical requirements.