The aim of the paper is twofold: firstly defining a method for object extraction from gray-level graphical images; secondly recognizing graphical objects in order to return an automatic description of a given complex drawing. The segmentation uses an histogram mode clustering, which groups the pixels by gray-level intensity in order to define a series of thresholds. A multithreshold method is developed to insert local properties in the multimodal hystogram and to realize an automatic threshold selection. The automatic detection of gray-level images, representing technical drawing, may be simplified by some given drawing criteria. In order to recognize the graphical objects a gool-driven approach is adopted. A structural model is then defined in which the domain knowledge is represented by a semantic network. Finally the semantic network knowledge is used to recognize part or set of parts of a technical drawing, according to a given strategy.