In the current science-based (as distinguished from ethics-based) debate on the use of nonhuman animals in biomedical and psychological research, scientific evaluation of animal models is a critical issue. Regrettably, much debate has focused instead on similarities and differences between the model and the modeled. Through a review of the proper understanding and use of the strategy of model-building in science, the paper shows that, whether based on theoretical or empirical grounds, the apples and oranges argument— that model and modelled are incommensurable — is not an adequate basis of evaluation. Continuing argument that relies heavily on similarities versus differences is unconstructive, as it is not a bottom-line consideration in the evaluation of a given model. What is critical is 1) whether we learn and 2) whether we improve treatment through the animal model research. A novel method of evaluation and its application to a set of animal models is presented.