Psychological testing can serve as a prototype on which to base a data-generated approach to instructional design. In "testing batteries" tests are used to predict achievement. In the proposed approach batteries of prescriptions would be used to produce achievement. In creating "test batteries" tests are selected for their capacity to tap into the valid variance associated with achievement. Individual tests are assembled into a battery on the basis of the size of their correlation with achievement. Cumulatively the individual tests selected make the best fit for predicting it (R[superscript 2]). For "prescription batteries" a prescription would first be nominated on the basis of rationales. It would then be selected for its capacity to tap into the valid variance associated with achievement. Batteries would be created by assembling multiple individual prescriptions which, cumulatively, are likely to produce achievement. Data serving as the basis for this approach would consist of correlations Current design models, which are postulated and not data-generated, are on vulnerable ground with respect to two methodological issues. They still need to demonstrate that they can be implemented in consistent fashion. There are identifiable conditions which can prevent this from occurring. Similarly there are identifiable conditions which can prevent models from being applied in valid ways. It is their uncertain status with respect to reliability and validity which raises questions about instructional design's current "science" status. An empirical, data-driven approach, one version of which is being proposed here, can overcome some of these obstacles and go some way in mitigating others. With reliability and validity demonstrated, instructional design may be put on a more secure path to becoming a prescriptive "science."