With the tide of artificial intelligence, we try to apply deep learning to understand 3D data. Point cloud is an important 3D data structure, which can accurately and directly reflect the real world. In this paper, we propose a simple and effective network, which is named PyramNet, suites for point cloud object classification and semantic segmentation in 3D scene. We design two new operators: Graph Embedding Module(GEM) and Pyramid Attention Network(PAN). Specifically, GEM projects point cloud onto the graph and practices the covariance matrix to explore the relationship between points, so as to improve the local feature expression ability of the model. PAN assigns some strong semantic features to each point to retain fine geometric features as much as possible. Furthermore, we provide extensive evaluation and analysis for the effectiveness of PyramNet. Empirically, we evaluate our model on ModelNet40, ShapeNet and S3DIS.