The "accordion effect" is an effect of language which allows us to describe one and the same thing more or less narrowly. Social capital has been conceived in terms of our access to institutional resources, but also in terms that extend to the levels of trust and related resources found in the social networks we are embedded in. The former conception is narrower, favoured for its specificity and analytical utility. The latter conception is broader, favoured for its acknowledgement of context, including the qualitative features of relations between individuals and within communities. These conceptions appear incompatible, but both have numerous adherents in educational research, and it is unclear whether either can be eliminated without some threat to the intelligibility and explanatory promise of social capital theory in an educational context. This raises hard questions about the domains, questions and methods that the social capital concept is best applied to. Should the concept prove resistant to any defensible specification, this will require a significant revision to the stock of conceptual resources available for understanding and explaining educational progress.