Abstract: Divergence as a degree of the difference between two data is widely used in the classification problems. In this paper, f-divergence, H-divergence and δ-divergence of the set-valued measures and non-additive set-valued measures are defined and discussed respectively. It is proved that H-divergence and δ-divergence satisfy the triangle inequality and symmetry by means of the set operations and partial ordering relations. Meanwhile, the necessary and sufficient conditions of Radon-Nikodym derivatives of the set-valued measures and non-additive set-valued measures are investigated respectively. Finally, some examples are given to illustrate the effectiveness of the definitions and results proposed in the article.