Social scientists doing fieldwork in humanitarian situations often face a dual imperative: research should be both academically sound and policy relevant. We argue that much of the current research on forced migration is based on unsound methodology, and that the data and subsequent policy conclusions are often flawed or ethically suspect. This paper identifies some key methodological and ethical problems confronting social scientists studying forced migrants or their hosts. These problems include non-representativeness and bias, issues arising from working in unfamiliar contexts including translation and the use of local researchers, and ethical dilemmas including security and confidentiality issues and whether researchers are doing enough to 'do no harm'. The second part of the paper reviews the authors' own efforts to conduct research on urban refugees in Johannesburg. It concludes that while there is no single 'best practice' for refugee research, refugee studies would advance its academic and policy relevance by more seriously considering methodological and ethical concerns.