Abstract. The notion of levels has been widely used in discussions of cognitive science, especially in discussions of the relation of connectionism to symbolic modeling of cognition. I argue that many of the notions of levels employed are problematic for this purpose, and develop an alternative notion grounded in the framework of mechanistic explanation. By considering the source of the analogies underlying both symbolic modeling and connectionist modeling, I argue that neither is likely to provide an adequate analysis of processes at the level at which cognitive theories attempt to function: One is drawn from too low a level, the other from too high a level. If there is a distinctly cognitive level, then we still need to determine what are the basic organizational principles at that level.Key words. Connectionism, symbol processing, levels of organization, reduction, mechanistic explanation.The recent attention given to connectionist, parallel distributed processing, or neural network models of cognition has raised a fundamental question about how these inquiries relate to other attempts to explain cognitive phenomenon. The use of the three different names sometimes interchangeably and sometimes distinctively revels that there is a fair amount of disagreement as to how to answer this question. The name neural networks suggests that these are models of actual neural systems. For some theorists, this implies that these models are, at best, tangentially relevant to the business of cognitive science. They may be useful for studying how structures in the brain function, and since cognitive activities are realized in the brain, they may be relevant for studying how cognitive activities are realized in the brain, but that is all. Some neuroscientists, especially those who adopt the term cognitive neuroscience for their pursuit, however, would insist that neuroscience is in the business of explaining cognitive functions. Thus, there are some investigators who would employ the term neural networks who take themselves to be engaged in explaining cognitive functions. For many of these theorists it is important to develop models of how the brain performs cognitive functions, and they are therefore quite skeptical of the research programs commonly associated with the term cognitive science which have not made faithfulness to neural mechanism primary.Other theorists, who are more likely to have come to such network models by way of psychology or artificial intelligence, tend to prefer the labels connectionism or parallel distributed processing 2. This choice of names reflects a different conception of the enterprise. For these theorists, network models are attractive only in part because of their similarity to networks of neurons. The attraction is more due to the fact that such networks provide useful frameworks for modeling