the complexities involved in policies that depend on decisions taken at the multiple and sometimes conflicting levels of a system of higher education. Based on the Swedish example, this article examines and criticizes some of the current models employed in analysing developments in this field. I shall argue that many of them suffer from a weak conceptual framework and that different models tend to emphasize different aspects of the data, not always in a complementary manner. In this paper I shall pay particular attention to five models or explicatory frameworks employed to analyse innovation and change in higher education. These are the elite, mass and universal model developed by Trow, the Rational Decision Model, an Incremental Model, a Political Model and, finally, what has been termed by Cohen, March & Olsen, the 'Garbage Can' Model. In conclusion, the paper suggests that a new framework is needed based upon what may be termed a 'Public Administration' Model. The empirical data with which this paper is concerned relate to the post-war development of the Swedish system of higher education. The raw material is almost inexhaustible and could concern the growth of student numbers, the expansion of various kinds of professional and administrative personnel, and the introduction of egalitarian admission criteria-all socially relevant; and there exist a number of corresponding choices as to what data to report. Some may be selected for their political relevance, others for their significance in descriptions of higher education systems, and yet others for their appropriateness in some historical description. Often, in discussions of this kind, problems of selection are left to be handled by 'implicit' criteria, or are subject to several different confusing principles. Here, however, the task of selecting empirical information is to be solved by means of a theoretical approach, without which there is no limit to the relevant facts to be reported, so that any other form of circumscription would be random or arbitrary. Far from being self-sufficient, data have to be interpreted, examined from various angles, and arranged into some kind of structure. Problems of iatelpietation are often handled through the application of firmly established inherited traditional concepts, which are always readily available but clouded by ambiguity and vagueness. A theoretical