This paper introduces SAGPool, a novel graph pooling method based on self-attention for graph neural networks (GNNs). The method aims to improve the performance of graph classification tasks by considering both node features and graph topology. Unlike existing pooling methods that primarily focus on graph topology, SAGPool uses self-attention mechanisms to select nodes for retention, leveraging graph convolution to calculate attention scores. This approach ensures that the pooling process is both efficient and effective, achieving superior performance on benchmark datasets with a reasonable number of parameters. The paper also discusses the experimental setup, including the evaluation of different pooling methods and the analysis of the proposed SAGPool method, highlighting its advantages over existing techniques.This paper introduces SAGPool, a novel graph pooling method based on self-attention for graph neural networks (GNNs). The method aims to improve the performance of graph classification tasks by considering both node features and graph topology. Unlike existing pooling methods that primarily focus on graph topology, SAGPool uses self-attention mechanisms to select nodes for retention, leveraging graph convolution to calculate attention scores. This approach ensures that the pooling process is both efficient and effective, achieving superior performance on benchmark datasets with a reasonable number of parameters. The paper also discusses the experimental setup, including the evaluation of different pooling methods and the analysis of the proposed SAGPool method, highlighting its advantages over existing techniques.