In the context of automated reassembling of manually torn document snippets contour based approaches are insufficient because snippets have the same rupture edges if more than one page is torn at the same time. Moreover jigsaw puzzling is np hard which requests for a grouping of document snippets beforehand such that the complexity and computational speed of reassembling is improved. Analyzing the visual content of document snippets renders the distinction of snippets with the same contours possible. In addition, a visual content extraction enables fine alignment of snippets with the same content and for grouping snippets. The document analysis approaches presented in this thesis are part of a combined reassembling which utilizes content and contour for the reconstruction of about 600 Million Stasi snippets. The ruling analysis classifies the supporting material into void, lined, and checked paper. If a ruling is detected, the lines are localized accurately which allows for snippet alignments. Snippets might have sparse visual content depending on the conscientiousness when tearing. Therefore a new word localization (the so-called Profile Box) is introduced which keeps a compact word representation while accounting for anticipated deformations such as a word's local skew. These word boxes are further classified into printed, manuscript, and non-text elements by means of Gradient Shape Features (GSF) which are designed newly for this task. The latter class allows for rejecting falsely binarized elements which improves the robustness in the presence of degraded or noisy documents. Finally, a layout analysis is performed that is based on a bottom-up approach to keep the element clustering flexible even if a global text structure is not present. Results on various publicly available databases show that the methodology is capable of being adopted to different document analysis scenarios. A synthetic database for ruling line removal is created and made publicly available which allows comparisons between the approach proposed and other state-of-the-art methodologies. The text classification is compared to other approaches by means of the PRImA benchmarking database and the Iam database, which is a handwriting database written by multiple authors. The methodology presented achieves the best results in both empirical evaluations. On real world Stasi snippets, the recognition rate is lower because of the heterogeneity and sparseness of content in the data. The layout analysis is additionally evaluated on the most recent Handwriting Segmentation Contests where it competes state-of-the-art methods and on a medieval database.