This paper combines the findings of a detailed corpus study and the results of a small-scale experiment in an attempt to support the idea that grammatical knowledge includes knowledge of usage patterns. The linguistic structures investigated are English not that sentences of the type Other people have friends. I have enemies. Not that I care (example taken from the British National Corpus). An in-depth investigation of 1,759 uses of this structure extracted from the British National Corpus reveals that neither their lexico-grammatical environments nor the pragmatic functions they serve are distributed in a random or unpredictable way, but rather pattern to an astonishing extent. This finding prompts the question as to whether the observed patterns are merely a matter of repeated usage driven by recurring pragmatic needs, or instead are represented as part of the linguistic knowledge of individual speakers, i.e., as a part of grammar. In order to answer this question a simple experiment was designed aiming to investigate whether there is a match between the data observed in the corpus (reflecting usage) and the knowledge-based behavior of individual speakers in an offline task (tapping into grammar). 30 native speakers of English participated in a test which put them under severe time pressure while they filled in gaps in short texts. Four gaps targeted uses of not that representing patterns with different frequencies in the corpus. The results of the experiment indicate that the participants were faster and found it easier to activate the structure not that for stimuli targeting frequent patterns than for those targeting very rare types of usage. This is interpreted as evidence for the assumption that frequency-related knowledge about lexico-grammatical and even pragmatic patterns is a part of grammatical knowledge.