Brain activity associated with directing attention to one of two possible sensory modalities was examined using high-density mapping of human event-related potentials. The deployment of selective attention was based on visually presented symbolic cue-words instructing subjects on a trial-by-trial basis, which sensory modality to attend. We measured the spatio-temporal pattern of activation in the approximately 1 second period between the cue-instruction and a subsequent compound auditory-visual imperative stimulus. This allowed us to assess the flow of processing across brain regions involved in deploying and sustaining inter-sensory selective attention, prior to the actual selective processing of the compound audio-visual target stimulus. Activity over frontal and parietal areas showed sensory specific increases in activation during the early part of the anticipatory period (~230 ms), probably representing the activation of fronto-parietal attentional deployment systems for top-down control of attention. In the later period preceding the arrival of the "to-be-attended" stimulus, sustained differential activity was seen over fronto-central regions and parieto-occipital regions, suggesting the maintenance of sensory-specific biased attentional states that would allow for subsequent selective processing. Although there was clear sensory biasing in this late sustained period, it was also clear that both sensory systems were being prepared during the cue-target period. These late sensory-specific biasing effects were also accompanied by sustained activations over frontal cortices that also showed both common and sensory specific activation patterns, suggesting that maintenance of the biased state includes top-down inputs from generators in frontal cortices, some of which are sensory-specific regions. These data support extensive interactions between sensory, parietal and frontal regions during processing of cue information, deployment of attention, and maintenance of the focus of attention in anticipation of impending attentionally relevant input.