We present a new variational method for mesh segmentation by fitting quadric surfaces. Each component of the resulting segmentation is represented by a general quadric surface (including plane as a special case). A novel energy function is defined to evaluate the quality of the segmentation, which combines both L 2 and L 2,1 metrics from a triangle to a quadric surface. The Lloyd iteration is used to minimize the energy function, which repeatedly interleaves between mesh partition and quadric surface fitting. We also integrate feature-based and simplification-based techniques in the segmentation framework, which greatly improve the performance. The advantages of our algorithm are demonstrated by comparing with the state-of-the-art methods.