Despite the ill-posedness of the clustering task, in the continuous setting a broad consensus is overall acknowledged in defining the concept of cluster. Conversely, a general notion of cluster remains controversial in the presence of categorical data. We propose a novel notion of cluster hinging on the twofold concept of high frequency and association between variables. The former concept, in fact, complies with the cluster notion described by the modal formulation of the clustering problem, which we take advantage of to borrow some operational tools to propose an operational procedure.