This chapter addresses the problem of ancient Arabic document segmentation. As ancient documents neither have a real physical structure nor logical one, the segmentation will be limited to textual area or to line extraction in the areas. Although this type of segmentation appears quite simple, its implementation remains a challenging task. This is due to the state of the old document where the image is of low quality, the lines are not straight, sinuous and connected. Given the failure of traditional methods, we proposed a method for line extraction in multi-oriented documents. The method is based on an image meshing that allows it to detect locally and safely the orientations. These orientations are then extended to larger areas. The orientation estimation uses the energy distribution of Cohen's class, more accurate than the projection method. Then, the method exploits the projection peaks to follow the connected components forming text lines. The approach ends with a final separation of connected lines, based on the exploitation of the morphology of terminal letters.