In this work we present a methodology oriented to domains with a weak structure (ill-domains) for, using inductive conceptual learning techniques (descriptive generalization) and classification, discover concepts through observations from the domain, and organize hierarchies with them, in order to, after expert validation, build knowledge bases. Some techniques for the improvement of the results in the classification step are used, like biasing using partial expert knowledge (classification rules or causal and structural dependencies between attributes) or delayed cluster assignation of objects.