The reasons why students from lower socio-economic groups are under-represented at high status universities are not yet entirely understood, but evidence suggests that part of the gap may be a consequence of differential choice of A-levels by social background. The Russell Group of universities has since 2011 published guidance on A-level subject choices, describing some A-levels as "facilitating" in that choosing these helps keep the largest number of Russell Group degree courses open. The numbers of students gaining AAB in facilitating subjects has subsequently been developed as a performance measure for individual schools and sixth form colleges, and, in aggregate, as a government Social Mobility Indicator. While it is clear that there is a gap between the proportions of students in maintained and private schools achieving this measure, there is little other work to date on how social background is related to the take-up of facilitating subjects, or to a more fine-grained categorization of all the large number of "non-facilitating" subjects. I develop a taxonomy of all 96 A-levels certified for English students in 2014/15 beyond the facilitating/non-facilitating dichotomy according to Russell Group members' expressed preferences, and use it to analyse the A-level subject choices of three A-level cohorts (2010-2012), using National Pupil Database data. I find that large differentials in A-level subject choice exist by social background, particularly for facilitating subjects. These differentials substantially disappear when GCSE attainment and subject choices are taken into account. Closing this choice gap at A-level is likely therefore to depend on reducing differentials in attainment and subject choice by social background at GCSE. The introduction of the EBacc may help with the GCSE subject choice element.