IThe interpretation of data requires a model of the data gathering process. The integration of data requires a classification system that extends across different data sets. This article discusses some of the difficulties involved in the interpretation and integration of psychological data, and presents a survey of some recent developments in theory construction and evaluation from related areas of methodology of science, information processing and computer theories. The basic behavioral data of psychology consist of descriptions of when who emitted what under which conditions. There are so many whens, whos, whats, and whichs that it is impossible to store them all in all their detail (though some psychologists are trying to do just that). The lack of storage facilities forces everyone to record only those special aspects that he considers important at the moment. Different observers often use different classification categories, and the same observer often changes his system from one experiment to the next. The absence of any over-all theory or methodology serves to perpetuate and even to justify this diversity. For that matter, it is after all exactly the job of an over-all theory to reduce a bewildering variety of data to manageable proportions: to provide a 'key' or code which unlocks significant patterns of data from the noisy confusion. And especially important are those theories which integrate data across different data sets, since that results in powerful generalizations which are more independent from particular aspects of the way observations and measurements are set up. The historical situation by the mid-sixties was such that relatively little progress had been made by psychologists to resolve these problems (Murphy, 1963; Sells, 1963; Toda and Shuford, 1964). The anti-theory attitude that had prevailed in psychology (Boring, 1963; Chapanis, 1963;Marx, 1963)had