The talk takes part in the wider conceptual engineering project that calls for conducting normative, and not merely descriptive research on representational devices, called 'concepts,' no matter if understood in a linguistic or psychological sense. I will discuss a novel, conceptual-engineering-friendly interpretation of the counterexample use of thought experiments. According to this interpretation, this method justifies a revision of a certain concept following our expectations towards its function. I will show why this approach avoids the empirical-based critique aimed at demonstrating the unreliability of the use of thought experiments on the basis of cultural and other demographic differences in verdicts on thought experiments, and how it can enrich the methodological toolkit of those interested in conducting normative work on concepts. Moreover, I will show, however, that within a theory of reference, the discussed method cannot be used to revise the concept of REFERENCE. The reason for this is that intuitions of extension in a theory of reference constitute a set of data that needs to be explained, whereas in most other philosophical theories, they could either be abandoned in favor of our expectations towards the scrutinized concept’s function or serve as a criterion for the adequacy of the proposed definition of the target concept. In sum, I will demonstrate how, with respect to many fields, thought experiments may serve to revise our representational devices, and I will argue that, at least with respect to the phenomenon of the reference relation, even conceptual engineering requires descriptively aimed research.