For three-dimensional mechanical design and reverse engineering in manufacturing, point clouds are obtained from some scanners before they are used to generate geometrical shapes in a design. However, original point clouds are poor in quality because of noise, incomplete, and non-uniform data samples. Simplification is an important step to generate a good result prior to polygonal meshes. Usually we cannot obtain uniform points using traditional cloud simplification methods. This paper proposes a new method for non-uniform points cloud simplification (NUPCS), which is based on affinity propagation clustering. Experiments are carried out for some data sets and results show that our proposed method can deliver good simplification performances.