Rainfall measures may be imperfect proxies for floods, given factors such as upstream water balance, proximity to rivers, and topography. We check the robustness of flooding-migration relationships by combining nationally-representative survey data with measures of flooding derived from weather stations, gridded products, and remote sensing tools. Linear probability models reveal that extreme flooding is negatively associated with out-migration. Rainfall-based proxies produce results qualitatively similar to those using the satellite-based measure of inundation, but only the latter is able to discern non-monotonic effects throughout the distribution. Moreover, estimates differ widely across areas, suggesting that households respond differently to rainfall and flooding.