A reoccuring problem in mining association rules is the selection of interesting association rules within the overall, and possibly huge set of extracted rules. The majority of previous work in this area relies on statistical methods for quality estimation and se-lection of association rules. However, strictly bottom-up approaches are oblivious of knowledge though knowledge may be available (e.g., provided by ontologies), and rule extraction may take advantage of it. In this paper, we conceive of the problem of selecting association rules as a classification task. A framework of a binary probabilistic classifier is introduced that uses ontologies in order to estimate whether and to which degree a rule expresses a mere taxonomic relationship. In so doing, selection of association rules (selection by elimination) is carried out by identifying and discarding trivial association rules.