We present a novel approach for modeling hair using level-of-detail representations. The set of representations include individual strands, hair clusters, and hair strips. They are represented using subdivision curves or surfaces, and have the same underlying base skeleton to maintain consistent high-level physical behavior when a transition between different levels-of-detail occurs. This framework supports automatic simplification of dynamic simulation, collision detection, and graphical rendering of animated hair. It also offers flexibility to balance between the overall performance and visual quality, and can be used to model and render different hairstyles. We have used these level-of-detail representations to animate various hairstyles and obtained noticeable performance improvement, with little loss in visual quality.