Recent advances in the development of animal models of human mental retardation (MR) syndromes offer exciting possibilities for understanding the pathogenic processes in these disorders. However, because MR is, by definition, a disruption in “cognition,” the potential offered by these models can only be fully realized if the attention devoted to the cognitive assessment of the animals is equal to that given to the genetic manipulation that created them. Accordingly, this paper provides guidelines for assessing cognitive function in animal models of human cognitive pathology, with an emphasis on MR syndromes. One of the major issues considered is task selection. Because different cognitive processes depend on different brain systems, the nature of the brain damage will determine the tasks that will be most sensitive in any given disorder. Tasks that are most sensitive to one disorder will often reveal no dysfunction in a different disorder. It is therefore imperative that task selection is guided by knowledge, or hypotheses, about (a) the neural systems disrupted in the target disorder; and/or (b) the specific cognitive abilities impaired in the target human syndrome. For example, a hallmark deficit in many MR syndromes is an impaired ability to transfer learning from one situation to another. Because this process has rarely been tested in animal models of MR, it is likely that the degree of impairment in the animals has been significantly underestimated. When devising tasks for animals based on the human data, however, there are dangers in developing analogous tasks, and even in using identical ones. These problems are discussed, along with potential solutions. A second major theme of the paper is that critical information can be gleaned by analyzing the details of subjects' performance, rather than by examination of success and failure rates alone. These types of in‐depth analyses can aid in specifying the nature of the impairment, and in illuminating the neural bases of the dysfunction. Examples of useful techniques for analyzing behavior and understanding brain‐behavior relationships are provided. © 1997 Wiley‐Liss, Inc.