Disasters are unpredictable and frequently lead to chaotic post-disaster situations, creating numerous methodologic challenges for the study of the mental health consequences of disasters. In this commentary, we expand on some of the issues addressed by Kessler and colleagues, largely focusing on the particular challenges of (a) defining, finding, and sampling populations of interest after disasters and (b) designing studies in ways that maximize the potential for valid inference. We discuss these challenges - drawing on specific examples - and suggest potential approaches to each that may be helpful as a guide for future work. We further suggest research directions that may be most helpful in moving the field forward.