Animal and human studies support multisensory interactions (MIs) early in the sensory processing stream. We assessed whether intermodal attention influenced early multisensory interactions. We used event-related potentials (ERPs) to investigate the latencies and scalp topographies at which an auditory-visual multisensory stimulus elicited different ERP amplitudes than the sum of auditory and visual stimuli presented alone. Participants selectively attended to the auditory, visual, or tactile modality in a trisensory intermodal attention task. With auditory attention, early (~20 ms) MIs were observed over frontal electrodes. Visual attention delayed the MI latency to 35-75 ms with more widespread effects over frontal, central, and posterior scalp electrodes. Attention also modulated longer latency MIs with auditory and tactile attention. MIs were observed over frontal electrodes at 100-­140 ms. MIs with visual attention were broadly distributed at 200-240 ms and current source density topographies revealed a posterior-contralateral distribution of the MI, significant at 140-150 ms and 210-220 ms. A centrally distributed and negative polarity MI ERP at 270-315 ms was apparent in all three attention conditions. These results indicate that multisensory interactions are modified by intermodal selective attention at multiple stages in the sensory processing stream.