Traditionally produced lexical resources for Serbo-Croatian are not suitable for automatic processing of contemporary Serbian. More specifically, the processes of structural derivation, although very productive in Serbian, are not presented in either monolingual or bilingual dictionaries in a systematic way. The morphological e-dictionary of Serbian was initially produced on the basis of traditional resources and as such reproduces the same flaws. In order to overcome them two solutions are possible. One is to put in the e-dictionary all the lemmas produced by structural derivation, no matter whether they are recorded in traditional dictionaries and confirmed in the corpus of contemporary Serbian, and then to assemble them explicitly in a complex lemma by an appropriate lexical graph. This approach, however, implies an overproduction of lemmas and the construction of such a graph for each complex lemma. Another approach is to extrapolate the missing lemmas using morphological grammars that model specific morphological processes and that are applied only to the entries already in the dictionary, which is checked by using the lexical constraints. It is demonstrated that such morphological grammars enable a precise classification of processes of structural derivation in a way similar to the classification of inflective phenomena.