Health surveys are a very important component of the epidemiology toolbox, and play a critical role in gauging population health, especially in developing countries. Research on health survey methods, however, is sparse. In particular, current sampling methods are not well adapted for certain 'difficult' settings, such as emergencies, remote regions without easily available sampling frames, hidden and vulnerable population groups, urban slums and populations living under strong political pressure. This special issue of Emerging Themes in Epidemiology is entirely devoted to survey methods in such settings, and builds upon a successful conference in London highlighting problems with current approaches and possible ways forward. Greater investment in research on health survey methods is needed and will have beneficial effects for populations in need.Health surveys are the stethoscope, thermometer and pressure gauge of global health. Measurement of the health-based Millennium Development Goals depends on large-scale surveys such as the Demographic and Health Surveys, Multiple Indicator Cluster Surveys, and Living Standard Measurement Surveys [1]. For most international health interventions, including preventive disease control, curative care, health system strengthening, and emergency relief, population surveys are necessary to monitor implementation. Surveys can also provide direct measures of health outcomes and impact at the population level, and highlight important differentials in exposures and/or disease risk within particular groups, thus providing a trigger for action.Despite the contribution that survey data can make to global health improvement, research to develop survey methods in difficult settings has largely stagnated over the past two decades. A mere handful of studies on this topic have been published. This may be because of a perception that surveys do not require the same sophistication and rigour as other types of studies, such as clinical trials. Yet surveys present a number of technical challenges, including the need to select representative samples, achieve adequate statistical precision and minimise bias in data collection.