Rendering highly complex models can be time and space prohibitive, and decimation is an important tool in providing simplifications. A decimated model may replace the original entirely or provide level-of-detail approximations. Here we present and evaluate, quantitatively and qualitatively, methods for rapidly decimating volumetric data defined on a tetrahedral grid. Results are compared using both direct volume rendering and isosurface rendering. A mass-based and a density-based decimation error metric are compared, and the mass-based metric is found to be superior. Grid surface vertices are decimated using a geometric error metric, as well as one of the data-based error metrics. Images produced using direct volume rendering and isosurface extraction on grids that are decimated approximately 80% are nearly indistinguishable from similar images using the non-decimated grids, and even at 95% decimation, the rendered images have few artifacts. Rendering speed-up depends upon the rendere...