In this paper we propose a new measure of the usefulness of a statistical model in terms of its relative ability to predict new data. The new measure, to be called GENO, combines ideas from Erev, Roth, Slonim, and Barron (2007), from the well-known AIC criterion for model selection, and from cross-validation. GENO depends on the nature of the data and the given sample size, reflecting the fact that the usefulness of a model depends on both the adequacy of the model as a description the data generation process, and on our ability to estimate the model\u2019s parameters. Our research was initially motivated by the study of data from experimental economics, and we also provide a detailed biological example.<\/p>\n\n