The research will focus on four outstanding problems in cognitive psychology, with overall direction and coherence supplied by continuing comparisons of memory storage and retrieval models with parallel distributed processing models: 1. An important form of classification learning is the development of skill at judging the probable causes of symptoms, as in medical diagnosis or trouble shooting of complex equipment. Such judgments depend both on similarities of new cases to previous ones and on frequencies of similar events in one's past experience. Analyses of simulated diagnostic situations will enable analysis of the way frequency information and similarity information combine in such judgments and the conditions for approaching optimal decision making. 2. Recognition is generally taken to be a more sensitive way of assessing information stored in memory then recall or other forms of testing. Past research has greatly clarified the basis for recognition judgments, but there has been little study of the underlying learning process. This deficit will be attacked by studies in which recognition tests are preceded by sequences of learning experiences, with the results interpreted in terms of adaptations of general models for classification learning. 3. How information acquired by learning fades from memory with time has long been studied, but the causes of retention loss are still not well understood. The view that a principal factor is the displacement of memories by new learning (or unlearning) that occurs between study and retention testing was undermined by numerous apparently negative research results. In this project, the hypothesis will be revived in the light of more effective models for learning and memory. Preliminary results indicate anew that new learning is an important factor in retention loss. 4. In decision situations such as gambles or choices between insurance packages, people normally make choices that tend to maximize probable returns; under some circumstances, however, conspicuous deviations from that type of rationality have been observed. One, termed the "sure thing" preference, is a tendency to choose an alternative that yields some certain payoff over another alternative whose outcome is uncertain but with a higher expected value. This project will investigate the nature of previous learning experiences that contribute to the presence or absence of the "sure thing" phenomenon and related deviations from optimal choice behavior. The expected result of this research is convergance on a model subsuming several of these types of behavior. Such a model would provide a firmer basis for our society's educational enterprise than models that only subsume one area of behavior.