This review aims to stimulate new ways of thinking about how to model depression in rats and mice. The article is founded on the premise that anthropomorphic inferences should be removed entirely from research involving rodents. The application of animal models to study depression over the past 30 years has been based largely on nonempirical and hence nonscientific assumptions about psychological states that probably do not exist and certainly cannot be measured in rodents. Such assumptions may have led to the misinterpretation of some behaviors, such as decreased locomotor activity or decreased sucrose consumption, as symptoms of depression in rats. Previous research has also overemphasized the causal role of stress in depression. After reviewing major features of several commonly employed models, this article challenges traditional concepts about validity. Models are first evaluated based on the goals of the research. Screening for potential antidepressant compounds requires little or no consideration of the validity of the model. Issues of validity become more critical when attempting to study the neurobiological basis of depression. The primary importance of face validity is emphasized, and the value of various behavioral measures is assessed based on how directly they resemble discrete behavioral symptoms seen in depressed humans. A "neurobehaviorally mechanistic" approach is described. This approach relies on formulating discrete, neurobiological hypotheses to explain individual symptoms rather than to explain collections of symptoms or the entire disorder. The approach thus relies on pragmatic measures of operationally well-defined behavioral variables. The review concludes with the proposal that understanding the neurobiological basis for individual symptoms will ultimately yield a better understanding of depression.