Descriptions of feeling states in nonhuman animals have relied on indirect evidence from empirical data. Assumptions that fish do not experience suffering lack evidence and in fact contradict a large body of indirect scientific evidence and ethical concern. Why should the burden of proof rest on those defending the hypothesis that fish feel pain and other discomfort? In this article I address this controversy and describe typical methodsand the problems associated with themto identify animal welfare (feeling-based, physiological, and behavioral approaches intended to demonstrate feelings and welfare states). Then I urge a shift in scientific focus from efforts to either identify an internal state of well-being or determine whether an organism suffers, to efforts to identify conditions that promote a "good state" for an animal (i.e., a state it would choose). For this approach, I discuss preference tests and their implications for scientific research, teaching, aquarism, and fishing.