Previous ERP studies have uncovered cross-modal interactions in endogenous spatial attention. Directing attention to one side to judge stimuli from one particular modality can modulate early modality-specific ERP components not only for that modality, but also for other currently irrelevant modalities. However, past studies could not determine whether the spatial focus of attention in the task-irrelevant secondary modality was similar to the primary modality, or was instead diffuse across one hemifield. Here, auditory or visual stimuli could appear at any one of four locations (two on each side). In different blocks, subjects judged stimuli at only one of these four locations, for an auditory (Experiment 1) or visual (Experiment 2) task. Early attentional modulations of visual and auditory ERPs were found for stimuli at the currently relevant location, compared with those at the irrelevant location within the same hemifield, thus demonstrating within-hemifield tuning of spatial attention. Crucially, this was found not only for the currently relevant modality, but also for the currently irrelevant modality. Moreover, these within-hemifield attention effects were statistically equivalent regardless of the task relevance of the modality, for both the auditory and visual ERP data. These results demonstrate that within-hemifield spatial attention for one task-relevant modality can transfer cross-modally to a task-irrelevant modality, consistent with spatial selection at a multimodal level of representation.