Observational studies, for example cohort and case-control studies in which patients are allocated treatment on a non-random basis, are thought by some investigators to be flawed. This view results from the fact that, unlike experimental methods (randomized controlled trials; RCTs), the results of such observational studies are vulnerable to confounding. However, this view assumes that satisfactory adjustment of differences in risk or prognosis between treatment groups is impossible and it ignores some of the limitations of RCTs. While many of the problems involved in conducting RCTs could be overcome, the practical implications for researchers and funding bodies mean this is often not possible. In such circumstances, observational studies offer an alternative to an absence of any scientific evidence. While making use of observational methods, researchers must acknowledge the associated limitations: the inevitable inability to take unknown confounders into account, non-blinding of practitioners and patients, and the inclusion of practitioners’ and patients’ treatment preferences.