We conducted two fMRI studies to investigate the sensitivity of delay-period activity to changes in memory load during a delayed-recognition task for faces. In Experiment 1, each trial began with the presentation of a memory array consisting of one, two, or three faces that lasted for 3 sec. A 15-sec delay period followed during which no stimuli were present. The delay interval concluded with a one-face probe to which subjects made a button press response indicating whether this face was part of the memory array. Experiment 2 was similar in design except that the delay period was lengthened to 24 sec, and the memory array consisted of only one or three faces. We hypothesized that memory maintenance processes that spanned the delay interval would be revealed by their sensitivity to memory load. Long delay intervals were employed to temporally dissociate phasic activity engendered by the memory array from sustained activity reflecting maintenance. Regions of interest (ROIs) were defined anatomically for the superior frontal gyri (SFG), middle frontal gyri (MFG), and inferior frontal gyri (IFG), intraparietal sulci (IPS), and fusiform gyri (FFG) on a subject-by-subject basis. The mean time course of activity was determined for all voxels within these regions and for that subset of voxels within each ROI that correlated significantly with an empirically determined reference waveform. In both experiments, memory load significantly influenced activation 6-9 sec following the onset of the memory array with larger amplitude responses for higher load levels. Responses were greatest within MFG, IPS, and FFG. In both experiments, however, these load-sensitive differences declined over successive time intervals and were no longer significant at the end of the delay interval. Although insensitive to our load manipulation, sustained activation was present at the conclusion of the delay interval within MFG and other prefrontal regions. IPS delay activity returned to prestimulus baseline levels prior to the end of the delay period in Experiment 2, but not in Experiment 1. Within FFG, delay activity returned to prestimulus baseline levels prior to the conclusion of the delay interval in both experiments. Thus, while phasic processes engendered by the memory array were strongly affected by memory load, no evidence for load-sensitive delay-spanning maintenance processes was obtained.