Non-metallic inclusions like oxides and sulfides influence the characteristics of tool steels. After manufacturing, the steel quality must be tested by visual observation of polished cut images in order to see if the steel can be used for further production steps. Since the human effort for this kind of quality inspection is very high the object of this work was to investigate whether or not it is possible to extract and classify steel inclusions by automated inspection. Therefore 47 images with a total of 131 inclusions (provided by the Böhler Edelstahl GmbH) have been pre-processed, analyzed and classified using different mathematical methodologies. Thereby state of the art operations as well as new developed methodologies have been applied. The classification of sulfides works very well. The separation of oxides turned out to be complex because oxides occur in three different but similar types. Finally, good classification results have been achieved for oxide types as well. As a concluding statement it is to say that a change to automated inspection and classification turned out to be possible. However it is advisable to further improve processing using more sample images of different inclusion types and different types of surface errors like holes and scratches.