This paper presents a novel approach to Conceptual Clustering in First Order Logic (FOL) which is based on the assumption that candidate clusters can be obtained by looking for frequent association patterns in data. The resulting method extends therefore the levelwise search method for frequent pattern discovery. It is guided by a reference concept to be refined and returns a directed acyclic graph of conceptual clusters, possibly overlapping, that are subconcepts of the reference one. The FOL fragment chosen is (mathcal{AL})-log, a hybrid language that merges the description logic (mathcal{ALC}) and the clausal logic Datalog. It allows the method to deal with both structural and relational data in a uniform manner and describe clusters determined by non-hierarchical relations between the reference concept and other concepts also occurring in the data. Preliminary results have been obtained on Datalog data extracted from the on-line CIA World Fact Book and enriched with a (mathcal{ALC}) knowledge base.