The birth of Connectionist Psychology (Feldman and Ballard 1982, McClelland & Rumelhart, 1986) has some resemblance with the one of modern Statistical Mechanics In both cases we deal with the introduction of a new, two-level (microscopic and macroscopic) description of the world, aimed to avoid the inconsistencies of an explanation of the experimental facts within the usual theoretical framework. These inconsistencies, in the case of Statistical Mechanics, derive from the impossibility of reconciling the phenomenological description of physical events, like the ones considered by classical thermodynamics, with the traditional newtonian dynamics. But in the case of Connectionist Psychology, they derive from the impossibility of reconciling the phenomenological description of cognitive phenomena with the one given by information-processing Psychology. In the same way as Statistical Mechanics, which does not negate the relevance of macroscopic behaviors but tries only to explain them in a simpler way through a more fundamental level of description, also Connectionist Psychology does not negate the relevance of symbolic information processing in cognitive phenomena but tries only to explain macroscopic cognitive behaviors through the cooperative action of a large number of simple cognitive microscopic units.