Previous approaches using quadric error metric allow fast and accurate geometric simplifications of meshes, but they don't consider rendering schemes and texture contents. A newly developed framework, image-driven simplification, uses image errors to decide which portions of model to simplify. The image-driven simplification is sensitive to rendering schemes and texture contents, and can produce simplified models with better visual quality. However, it can not deal with some models with highly complex visibility. In this paper, we introduce the rendering error metric and combine it with the quadric error metric. Benefits of using rendering error metric include higher quality simplified meshes, better preserved geometric and color boundaries, and simplification that is sensitive to the content of a texture. In contrast with the original image-driven simplification algorithm, our algorithm samples the model locally per edge collapse to compute the rendering error metric and is independent of specific viewpoints, so it can deal with any kind of models, even highly self-occluded models. We demonstrate the efficiency of the algorithm with a variety of meshes with normal/color/texture attributes.