This paper presents a novel dataset for hand gesture estimation. As we know, there lacks a well-designed hand gesture database so far, which greatly impedes the development of distinctive hand gesture estimation. In order to facilitate the research of hand gesture estimation, we build a new dataset of hand gestures. Our dataset contains about 5K labeled and 1K unlabeled single hand images. All the images are separated into 50 varieties according to the types of hand gestures. For each type of hand gesture, occlusion, light, shadow and background are all considered. We hope that the novel dataset would promote the development of distinctive hand gesture estimation from images.