This paper argues that grammatical theorizing and linguistic typologizing must go hand in hand and that rare typological features play a central role in the interaction of typology and theory. The paper is organized as follows. Section 2 discusses a sampling method that (compared to other sampling methods) offers the best chance of finding and identifying rara or rarissima (features and properties attested in very few languages), because the method in question is designed to generate samples that display the greatest possible linguistic variety. Section 3 discusses some relevant details about Functional (Discourse) Grammar and sections 4 and 5 are concerned with the crucial role of rara both in theory driven data collection and in data driven theory building.