Several models of judgment propose that people struggle with absolute judgments and instead represent options on the basis of their relative standing. This leads to a conundrum when people make judgments from memory: They may encode an option’s ordinal rank relative to the surrounding options but later observe a different distribution of options. Do people update their representations when making judgments from memory, or do they maintain their representations based on the initial encoding? In three studies, we found that people making memory-based judgments rely on a stimulus’s relative standing in the distribution at the time of encoding rather than attending to absolute quality or updating the stimulus’s ordinal ranking in light of the distribution at the time of the later judgment.