Résumé : Existing document decomposition models fail in well separating the zone areas of printed text and handwritten text when they are close or even touching each other. This paper presents a simple and robust algorithm to filter out the printed content in a mixed document. Following the traditional bottom-up approach, the printed text candidates are extracted and detected in the connected component level. Then the relative spatial relation and windowbased filter help providing useful information to decide the components to be removed finally. Results with manually cropped signature blocks that contain extraneous printed text show that over 85% of printed text components are removed while preserving handwritten content.