This paper presents a text line segmentation method for printed or handwritten historical Arabic documents. Documents are first classified into 2 classes using a K-means scheme. These classes correspond to document complexity (easy or not eas