Electoral surveys conducted as part of the regular series of British Election Studies have used a stratified, clustered sampling design. This is constructed to ensure a nationally representative sample of voters (after weighting) but does not necessarily ensure a representative sample of the different social areas within the country. Much recent work has indicated the important role of local social and electoral milieux in the structuring of electoral behaviour but these are not reflected in the sampling design. An analysis of the geography of the face-to-face pre-election component of the 2005 BES shows not only that it was unrepresentative of certain types of area (defined using bespoke neighbourhood data) but also that it was less representative of such areas than the two (larger) internet samples (pre- and post-election) also undertaken as part of the 2005 BES. This suggests the need to reconsider the nature of the sample designs (even the entire survey methodology) to be deployed at future election studies in Great Britain.