Functional magnetic resonance imaging is routinely used to localize brain function, with multiple brain scans averaged together to reveal activation volumes. In this study, we examine the seldom-studied effect of multiple scan averaging on the extent of activation volume. Using restricted visual field stimulation, we obtained a large number of scan repetitions and analyzed changes in activation volume with progressively increased averaging and across single scans. Activation volume increased monotonically with averaging and failed to asymptote when as many as 22 scans were averaged together. Expansions in the spatial extent of activation were not random; rather, they were centered about activation loci that appear with little or no averaging. Using empirical and simulated data, changes with averaging in activation volumes and cross correlation coefficient distributions revealed the presence of considerably more activated voxels than commonly surmised. Many voxels have low SNR and remain undetected without extensive averaging. The primary source of such voxels was not downstream venous drainage since there was no significant and consistent delay difference between voxels activated at different averaging levels. Voxels with low SNR may reflect a diffuse subthreshold activity centered about spiking neurons, dephasing gradients from distal veins, or simply a blood flow response extending beyond the locus of neuronal firing. Across single scans, as much as twofold changes in activation volume were observed. These changes were not correlated with the order of scan acquisition, subject task performance, or signal and noise properties of activated voxels. Instead, they may reflect subtle changes between overlapping noise and signal frequency components.