Halliday (2008: 113-4):
System and instance are the two observational poles: observers, we may position ourselves up close, as instance observers, or else at a distance, as system observers. The two vantage points raise different questions about language. But the phenomena we are trying to explain may be just those whose regularities are found in some intermediate region along the instantiation cline, so that it is not all obvious which position we should be observing from. … [Functional variation] appears either instantially in the form of “text types” or systemically as variation in “register”. If we find systematic variation in the meanings selected in a number of specimens of discourse, we might group together those instances that are alike in certain respects, in their grammar or in their text structure, and recognise a set of text types. But if want to explain such variation, we shift our stance to that of the system observer and recognise specific sub-systems [i.e. registers] that are motivated “from above” — from the context, associated with some culturally defined forms of social practice.