Point cloud sampling is a less explored research topic for this datarepresentation. The most commonly used sampling methods are still classicalrandom sampling and farthest point sampling. With the development of neuralnetworks, various methods have been proposed to sample point clouds in atask-based learning manner. However, these methods are mostly generative-based,rather than selecting points directly using mathematical statistics. Inspiredby the Canny edge detection algorithm for images and with the help of theattention mechanism, this paper proposes a non-generative Attention-based Pointcloud Edge Sampling method (APES), which captures salient points in the pointcloud outline. Both qualitative and quantitative experimental results show thesuperior performance of our sampling method on common benchmark tasks.