In this paper we discuss our approach to two areas of learning that have tended to be kept separate in the past, viz. proficiency and research-oriented descriptive heuristics. In proficiency courses the aim is for L2 students to acquire an advanced competence in English, while in research seminars students have to learn a methodology for language description. We will argue that these two aims can be integrated with each other by involving students throughout the curriculum in progressively more advanced forms of pattern discovery in English language corpora. We will illustrate this interaction between research and proficiency by means of the case study of ‘size nouns’. Brems (2003) showed through corpus-driven analysis that nouns such as bunch, load(s), pile(s), heap(s) have, besides literal uses as heads, also grammaticalized quantifier uses. This research informed the unit on size nouns in COLLEX, a corpus-based learning environment being developed by us, which aims at letting students acquire frequent and communicatively useful lexicogrammatical patterns. This case study of size nouns also illustrates the ongoing incorporation of results from research seminars in the higher years into the COLLEX proficiency resource offered to first year students.