We explore the determination of the optimal stimuli ensemble of a cortical network on the basis of numerical simulations of a local recurrent network. The optimal stimulus ensemble is determined with the Blahut-Arimoto algorithm using indices of population activity (multiunit activity and local field potentials). We show that the optimal stimulus ensemble is characterized by a distribution where the stimulus which induces a change in the dynamics of the network is highly represented. These results suggests that bifurcation points in the networks dynamic are highly informative and were in accordance with preliminary observation of electrophysiological recordings.