When faced with Duhemian problems of underdetermination, scientific method suggests neither a circumvention of such difficulties via the uncritical acceptance of background assumptions, nor the employment of epistemically unsatisfying subjectivist models of rational retainment. Instead, scientists are challenged to attack problems of underdetermination 'head-on', through a careful analysis of the severity of the testing procedures responsible for the production and modeling of their anomalous data. Researchers faced with the task of explaining empirical anomalies, employ a number of diverse and clever experimental techniques designed to cut through the Duhemian mists, and account for potential sources of error that might weaken an otherwise warranted inference. In lieu of such progressive experimental procedures, scientists try to identify the actual inferential work that an existing experiment is capable of providing so as to avoid ascribing to its output more discriminative power than it is rightfully due. We argue that the various strategies adopted by researchers involved in the testing of Bell's inequality, are well represented by Mayo's error-statistical notion of scientific evidence. In particular, an acceptance of her stringent demand for the output of severe tests to stand at the basis of rational inference, helps to explain the methodological reactions expressed by scientists in response to the loopholes that plagued the early Bell experiments performed by Alain Aspect et al.. At the same time, we argue as a counterpoint, that these very reactions present a challenge for 'top-down' approaches to Duhem's problem. / Master of Arts