This paper attempts to investigate whether we could in any meaningful way say of predicate expressions (whether monadic or relational) that they refer and, if they refer, what their reference could be. In semantics based on model theory (e.g., of Carnap's state-descriptions or Hintikka's model-sets), there are entities to which predicate expressions correspond (Carnap's attributes, Hintikka's predicates). What kind of entities are they?