The aim of this article is to establish the theoretical basis of the semantic annotation of the CESS-ESP and CESS-CAT corpora, by means of assigning to each predicate a Lexicosemantic Structure. The lexicosemantic patterns determine the argument structure, the correspondence between the arguments and the verb, and the diathesis alternations. The semantic annotation is carried out semi-automatically. As a result, a classification of predicates will be obtained and generalizations on the behavior of predicates will be drawn.