IntroductionSocial evaluations—the judgments and impressions we form about others—are deeply rooted in memories that integrate perceptual cues (e.g., facial features) and semantic knowledge (e.g., personality traits)1,2. These evaluations are often biased by heuristics such as the attractiveness halo effect, where individuals who are generally perceived as unattractive are systematically perceived as less competent, trustworthy, and likable3,4. Previous bias reduction interventions have relied on conscious and effortful strategies, such as counter-stereotype training. However, attitudes often resist deliberate modification due to defensive reactions5. A critical challenge, therefore, lies in identifying pathways through which social evaluations can be modified without engaging conscious control mechanisms.Sleep, particularly slow-wave sleep (SWS), provides a unique neurobiological state for modifying social evaluations because of its role in memory reprocessing and affective integration. During SWS, hippocampal-neocortical interactions based on spindle-oscillation coupling facilitate the selective strengthening and updating of associative memories, including affective valences attached to social stimuli6,7,8. While previous research has shown that sleep learning and memory reactivation can influence subsequent memory representations and behavior9,10,11, it remains unclear whether these processes specifically affect the updating of social evaluations during sleep. Pioneering work by Hu et al. (2015) proposed that targeted memory reactivation (TMR) during sleep could enhance pre-sleep anti-bias training, although subsequent replication attempts have yielded mixed results12,13. While these findings remain debated, they highlight the potential application of sleep-mediated interventions for modulating implicit associations. Since social evaluations are rooted in the associative memories that bind individuals with affective attributes, remembering more negative traits or having negative memories of a person’s facial appearance can lead to negative evaluations. In fact, the facial appearances contained in our memories, i.e., mental facial representations, can bias our perception and evaluation and are closely correlated with stigma and prejudice14,15. This raises a fundamental question: can sleep be leveraged to reconstruct social evaluations through associative learning?Our study aimed to address this gap by investigating whether social evaluations could be improved during sleep by linking acoustically cued names of individuals who were generally perceived as unattractive to novel positive traits during SWS. While modifying social evaluations during wakefulness may be more economical and cost-effective, this process often involves controlled cognitive processes that may conflict with preexisting perceptions or attitudes. For example, conscious efforts to suppress biases can paradoxically reinforce them through ironic rebound effects5. In contrast, sleep-mediated updating can bypass conscious resistance and may be more effective than updating participants while they are awake. Indeed, empirical studies have demonstrated that sleep-specific interventions—such as targeted memory reactivation (TMR)—produce unique behavioral effects that are not replicable when applied to participants who are awake13. This study tested the hypothesis that sleep-mediated associative learning can counteract attractiveness-based stigma by linking the acoustically cued names of individuals generally perceived as unattractive to novel positive traits during SWS. To assess social evaluations, we focused primarily on facial mental representations and examined social attitudes and the memory of personality traits. On the basis of previous sleep and TMR research, we hypothesized that, upon exposure to acoustic name+positive trait pairs during SWS, participants would form a more favorable mental representation of the cued individual and that their attitudes and memory biases toward that individual would become more positive (an overview of the study is presented in Fig. 1). At a broader level, this research could inform initiatives aiming to diminish stigma and prejudice toward various populations (e.g., mental illness, stigmatized groups, or minorities).Fig. 1: Overview of the study.The study consisted of four different phases. During the encoding phase, participants learned two faces with 10 personality trait descriptions. After encoding the faces and personality traits, the memory of those descriptions was tested, and an adapted implicit attitude test (IAT) was conducted. During SWS sleep, name+positive trait pairs were presented to the sleeping participants. After sleeping, participants performed post-sleeping tests [including mental facial representations, memory bias tests, and implicit (measured by the IAT) and explicit (measured by the ultimatum game and explicit preference rating test) attitudes]. Note. All facial images used in this study were obtained from the Tsinghua University Facial Expression Database36, and the original database confirms that written informed consent was obtained from all facial models for the use of their images in scientific research, including publications.Full size imageResultsPresenting acoustic name+positive trait pairs during sleep improved the attractiveness of mental facial representationsAs shown in Fig. 2I, the Linear Mixed Model (LMM) revealed that the main effect of the condition was significant (F(1, 8984) = 107.07, p