We investigate the limitations on the ability to detect when a target has changed, using Gabor targets as simple quantifiable stimuli. Using a partial report technique to equalise response variables, we show that the log of the Weber fraction for detecting a spatial frequency change is proportional to the log of the number of targets, with a set-size effect that is greater than that reported for visual search. This is not a simple perceptual limitation, because pre-cueing a single target out of four restores performance to the level found when only one target is present. It is argued that the primary limitation on performance is the division of attention across multiple targets, rather than decay within visual memory. However in a simplified change detection experiment without cueing, where only one target of the set changed, not only was the set size effect still larger, but it was greater at 2000 msec ISI than at 250 msec ISI, indicating a possible memory component. The steepness of the set size effects obtained suggests that even moderate complexity of a stimulus in terms of number of component objects can overload attentional processes, suggesting a possible low-level mechanism for change blindness.