The general goal of the present paper is to demonstrate how cross-linguistic (contrastive) data can broaden the perspective in cognitive linguistic research on metonymy, which may raise a host of questions calling for a revision of some widely accepted views. A more specific, methodological goal is to show how the introspection-driven research and the authentic-data driven research in cognitive linguistics can feed into each other in a cyclic way (cf. Kertesz and Rakosi, 2008b, p. 214ff). We highlight the role of contrastive research in the interaction between the introspection-driven and the authentic-data driven research in cognitive linguistics on two case studies. The first study is on the referential metonymy of the capital-for-government type. The second case study deals with the illocutionary metonymy motivating a range of constructions that realize the instructional speech act in cooking recipes. The two case studies have a number of things in common. In both of them the starting point is the question about the universality of a given metonymy. Similarly, in both studies we look for the motivation for the observed differences in the (non-)application of a given metonymy in a cross-linguistic perspective. After the analysis is extended to include a number of languages, the apparent impression in both studies is that the contrastive facts are ultimately motivated by some structural factors. Pursuing the analysis further, asking ourselves about the functional-cognitive background of the structural facts, we arrive at a deeper sort of motivation. (Contains 4 figures and 6 tables.)