This research is an attempt to apply a new kind of computational model, connectionist or parallel-distributed-processing models, to a number of complex, higher-level cognitive phenomena. These models are important, because a number of investigators claim that they are anatomically and physiologically more plausible than other sorts of models, and it is important to see how far they can go in accounting for the behavior of humans in the sorts of cognitive experiments that psychologists typically run. The first set of phenomena, involving attention and configural learning, includes the observation that over the course of discrimination learning, people come increasingly to "pay attention to" relevant cues, those that predict validly the correct response, and to ignore irrelevant cues. The second set of phenomena include generalization enhancement and distributed representations. The third set of phenomena involves representation issues, the issue of how different featural representations of stimuli should be mapped onto connectionist models. All of these issues will be approached with a combination of modeling and experimentation.