The authors present a multiresolution framework, called Multi-Tetra framework, that approximates volume data with different levels-of-detail tetrahedra. The framework is generated through a recursive subdivision of the volume data and is represented by binary trees. Instead of using a certain level of the Multi-Tetra framework for approximation, an error-based model (EBM) is generated by recursively fusing a sequence of tetrahedra from different levels of the Multi-Tetra framework. The EBM significantly reduces the number of voxels required to model an object, while preserving the original topology. The approach provides continuous distribution of rendered intensity or generated isosurfaces along boundaries of different levels-of-detail thus solving the crack problem. The model supports typical rendering approaches, such as marching cubes, direct volume projection, and splatting. Experimental results demonstrate the strengths of the approach.