Dental model repair is a critical stage in the orthodontic treatment. The repaired teeth need to preserve the original morphological features, and the algorithm is required to be automatic for saving time and avoiding uncertainty caused by human handling. This paper presents an efficient and automatic algorithm to repair the tooth models' large holes. Our algorithm consists of three main steps: model's principal axis determination, model's contours processing (generation and repair), and hole surface reconstruction and fairing. We propose a novel LRBC (Linear Regression on Boundaries' Centers) algorithm to determine the model's principal axis. After generating contours based on the principal axis, we adopt point-normal cubic B-spline curve interpolation to repair each broken contour line. At last, we reconstruct holes' point-sets by constrained Delaunay triangulation, and smooth the patches with a modified Laplacian operator. Experiments on different dental models demonstrate that our algorithm can not only achieve a result with a high morphological similarity and a low error, but also satisfy the requirements of actual orthodontic applications.