A discrete model of an ensemble of identical stochastic integrate–and–fire neurons is used to study the patterns of activity in populations of neurons that exchange excitatory messages. In a regime with small interactions among the units, the effect of the message exchange is to reduce the dispersion of the firing period of the individual neurons. In a strong interaction regime, a number of activity clusters emerge in the ensemble. Neurons in each cluster fire periodically and in synchrony with each other. The number of these self–sustained firing states characterized by distinct firing patterns towards which the network can evolve is very large. Because of their stability with respect to intrinsic fluctuations in the dynamics of the stochastic neurons, these states could, in principle, be used to encode and process large amounts of information.