Previous ERP studies focusing on supramodal spatial selection effects with visual and auditory stimuli regularly used block wise manipulations of focused attention and frequently employed a between-subjects design in which participants had to attend to either visual or auditory stimuli. A potential disadvantage of a design in which only visual or auditory stimuli are relevant is a strong suppression of any modality-irrelevant stimulation, thereby potentially underestimating supramodal spatial selection effects. In the current study, participants were instructed to attend to either the left or right and to either auditory or visual stimuli. They had to press a button when a deviant target of the relevant modality was presented on the relevant side (5% of the stimuli). The instruction varied on a trial-by-trial basis, and was indicated by a directional cue accompanied with a word indicating the relevant modality. A train of four stimuli, separated by 1 s, was subsequently presented, of which the location and modality varied randomly. ERPs showed not only supramodal but also modality-specific effects within the cue-target interval preceding the four stimuli. The P1 component for visual stimuli and N1 component for auditory stimuli were enlarged when the stimuli occurred on the to-be-attended side, irrespective from the relevance of the visual or auditory modality, suggesting that attentional selection shifts from an initial supramodal to a modality-specific level.