This study investigates the computational processes involved in the representation of numerical quantity by means of simulations with biologically plausible artificial neural networks. The main aim was to establish which kind of "internal representation " of numerosity emerges in a network that learns to enumerate a set of objects through an unsupervised learning procedure. Simulations show that the network develops a level of representation organized according to the cardinality principle. This result supports the assumptions of a previous models of numerical representation.