Dance is the real cultural wealth of any nation and dance media is the source for learning the dance steps by the present and future generation of students. Dance videos exhibit rich semantics and hence, the semantic modeling of the dance video objects is very crucial and complex. This paper presents a Dance Video Semantics Model (DVSM) in order to model the dance video objects at multiple granularity levels, identified by the components of the accompanying song. This model incorporates implicit, formal and power semantics of the dance videos. It also introduces a new entity type called, Agent, to represent the spatiotemporal actions of the actors which specify the contextual roles. The instantiations of the model are expressed as directed acyclic graphs.