We demonstrate that, in a parallel array of model neurons, the optimizing influence of internal noise on the global information is far greater than that reported for a single neuron. In particular, stochastic resonance (SR) effects, that optimize information transmission, occur independent of stimulus level or the setting of the neural threshold. We further show that adjusting the threshold to maximize information transmission does not remove SR effects. Consequently, and in contrast to a single neuron, in neuronal arrays noise appears to be an essential element of an optimal coding strategy.