We examined effects of trial averaging upon spatial extent, spatial topography, and temporal properties of fMRI activation. Two subjects participated in an event-related visual stimulation design. There was an exponential relation between number of trials and spatial extent, such that additional trials identified, on average, a constant proportion of the remaining voxels. At values typical of fMRI experimentation (e.g. 50 trials) only about 50% of eventually active voxels were significant; asymptotic values were approached by 150 trials. The variability of the estimated hemodynamic response decreased with signal averaging, becoming stable across samples of > or = 25 trials. Therefore, group or condition differences may result from differences in voxelwise noise exacerbated by averaging small numbers of trials.