Reconstructing media research so that generalizable knowledge can be produced requires an increased emphasis on attributes of media which interact with individual differences to effect learning. In order to accomplish this, media researchers must make a distinction between research "with" media and research "on" media. Privious efforts to systematize the selection of media for instruction by developing taxonomies generally have not been derived from previous research; they also do not draw on previous efforts to organize or conceptualize individual differences in learners. An alternative approach to the structuring and testing of media attribute taxonomies might list attributes to be validated by forming a three-dimensional matrix of subjects, behaviors, and attributes. The matrix is then collapsed across each of the three factors in turn and intercorrelations factored. By collapsing across subjects, for example, it is possible to compute the correlations between all the pairs of media attributes on specific behaviors and factor analyze the intercorrelation matrix. Factors would represent functionally similar media attributes; i.e., clusters of attributes which tend to elicit the same behavior. The result would form the basis of a media taxonomy for research purposes. (Author/WCM)