Harry Green¹ (), Chaoxiong Ye, Lijing Guo, Srihitha Akula, Alena Zheng, Heidi Lin, Nandani Sharma, Weizhen Xie; ¹Department of Psychology, University of Maryland, College Park, MD, USA, ²Department of Psychology, University of Jyväskylä, Finland

Visual recognition memory shows striking consistency across observers, a pattern often attributed to stimulus memorability – an intrinsic property of images predicting how likely they can be recognized. Yet, whether this consistency reflects only stimulus-driven factors or also stable individual differences remains unresolved. We examined this question using a continuous recognition paradigm with faces from the 10k U.S. Adult Faces Database, while assessing participants’ individual differences in response criteria during memory-based decisions. Across two samples (n = 146, U.S.; n = 240, China), participants viewed the same set of 160 critical faces that repeated only after longer delays ranging from 30s up to 40min, intermixed with filler images and vigilance repeats within 2s to 30s. In both samples, participants correctly responded to the same subset of repeated faces even under randomized test orders. This consistency was strongly predicted by established memorability scores, confirming a robust stimulus-driven component. However, participants’ correct detection of repetition systematically varied based on their response criteria. Individuals with more liberal response criteria, such as those with lower self-confidence and high psychosis traits, showed inflated rates of correct repetition detection. Their apparent “better memory” may reflect decisional bias rather than superior mnemonic encoding or higher stimulus memorability. These findings demonstrate that cross-observer consistency in visual recognition arises from the joint influence of stimulus-driven memorability and individual differences in response criteria. Conventional memorability estimates based on correct repetition detection rate (i.e., hit rate) is not solely a property of the images, but also a function of who is making the recognition decision. Together, these results underscore the need to account for individual differences in observer traits when interpreting “universal” patterns of visual recognition, and highlight that computer vision models trained exclusively on population-level hit rate to predict visual memorability may inherit systematic biases if individual differences are ignored.