We used event-related potentials (ERPs) to evaluate the role of attention in the integration of visual and auditory features of multisensory objects. This was done by contrasting the ERPs to multisensory stimuli (AV) to the sum of the ERPs to the corresponding auditory-only (A) and visual-only (V) stimuli [i.e., AV vs. (A + V)]. V, A, and VA stimuli were presented in random order to the left and right hemispaces. Subjects attended to a designated side to detect infrequent target stimuli in either modality there. The focus of this report is on the ERPs to the standard (i.e., nontarget) stimuli. We used rapid variable stimulus onset asynchronies (350-650 msec) to mitigate anticipatory activity and included "no-stim" trials to estimate and remove ERP overlap from residual anticipatory processes and from adjacent stimuli in the sequence. Spatial attention effects on the processing of the unisensory stimuli consisted of a modulation of visual P1 and N1 components (at 90-130 msec and 160-200 msec, respectively) and of the auditory N1 and processing negativity (100-200 msec). Attended versus unattended multisensory ERPs elicited a combination of these effects. Multisensory integration effects consisted of an initial frontal positivity around 100 msec that was larger for attended stimuli. This was followed by three phases of centro-medially distributed effects of integration and/or attention beginning at around 160 msec, and peaking at 190 (scalp positivity), 250 (negativity), and 300-500 msec (positivity) after stimulus onset. These integration effects were larger in amplitude for attended than for unattended stimuli, providing neural evidence that attention can modulate multisensory-integration processes at multiple stages.