Automatic Text Categorization (ATC) is an important task in the field of Information Access. The prevailing approach to ATC is making use of a a collection of prelabeled texts for the induction of a document classifier through learning methods. With the increasing availability of lexical resources in electronic form (including Lexical Databases (LDBs), Machine Readable Dictionaries, etc.), there is an interesting opportunity for the integration of them in learning-based ATC. In this paper, we present an approach to the integration of lexical knowledge extracted from the LDB WordNet in learning-based ATC, based on Stacked Generalization (SG). The method we suggest is based on combining the lexical knowledge extracted from the LDB interpreted as a classifier with a learning-based classifier, through SG. We have performed experiments which results show that the ideas we describe are promising and deserve further investigation.