Deverbal nominalizations constitute a rich source of semantic information. To have it recognized may be very useful for different NLP tasks. In this paper we present the nominal lexicon AnCora-Nom, which consists of 817 lexical entries of deverbal nouns, and a series of experiments based on machine-learning techniques. These experiments allow us to evaluate positively the consistency of annotated data in AnCora-Nom, and to detect the most relevant features for the denotative distinction between event and result nominalizations. Furthermore, with these experiments the foundations of an automatic classification system of the deverbal nominalizations according to their denotation are laid.