A large literature following Ruhm (2000) suggests that mortality falls during recessions and rises during booms. This relationship, however, tends to be analyzed within a panel-data framework that implicitly assumes either that local economic shocks do not induce migration, or that insofar as they do, these movements are accurately reflected in intercensal population estimates. In this paper, we argue that unobserved migratory responses have the potential to bias results. To study the extent of this bias, we draw on two natural experiments: the recession in cotton textile-producing regions of Britain during the U.S. Civil War, and the Appalachian coal boom that followed the OPEC oil embargo in the 1970s. In both settings, we find evidence of a substantial migration response. Furthermore, we show that estimates of the business cycle-mortality relationship obtained using the standard approach are highly sensitive to assumptions about both the accuracy of interpolated population values and the short-run relationship between population and mortality. We also show that control regions may be indirectly affected by migration into or away from treatment regions, leading to unobserved treatment spillovers. Together, our findings suggest that, when left unaddressed, large migratory responses can meaningfully undermine inference. Once we adjust for migration, we find no evidence that the coal boom substantially affected mortality, and we find that mortality increased during the cotton recession.