This study examines a residential sampling technique that was used to survey refugees in inner-city Johannesburg. The survey’s sampling framework assumed that, once displaced populations had settled, they could be a) adequately categorized according to residential sampling frames; and b) readily accessed in their homes. The paper explores the theoretical and practical limits of these assumptions. It shows how the particularly volatile relationship between respondents and the urban landscape made it difficult to generate a representative sample. The paper uses these findings to call for sampling procedures that are better suited to forced migrants experiences of urban space.