Experiments on short-term perceptual memory for elemental visual attributes, such as contrast, motion, orientation, and spatial frequency, have relied on a delayed discrimination technique in which the subject compares two stimuli presented at different points in time and memory is indexed by discrimination thresholds measured for the different time intervals between reference and test. In a variant of this procedure, used in experiments on long-term memory, the presentation of a single reference is followed by a memory test that combines two-alternative forced-choice decisions with the method of constant stimuli. With this procedure, it is impossible to distinguish the effects of criterion-setting processes and memory on performance, but this confound can be eliminated by testing many subjects and having each subject make a single decision. The resulting “group thresholds” are stable across time intervals of 24 hr, confirming previous findings of high-fidelity storage in the long-term memory range.