Emerging empirical evidence has demonstrated that visual working memory (VWM) representations are not totally independently stored, but rather undergo interactive influences from concurrently maintained information. When multiple items are similar to each other, meaning an individual perceives the distance between them in feature space to be small, their memory representations become distorted. These mnemonic distortions manifest two distinct directional effects: (1) attraction effects, characterized by a systematic bias of memorized items toward the features of neighboring representations (Brady and Alvarez, 2011, Chunharas et al., 2022, Lively et al., 2021, Son et al., 2020, Utochkin and Brady, 2020); and (2) repulsion effects, where stored representations exhibit directional shifts away from other memory items (Bae and Luck, 2017, Chunharas et al., 2022, Czoschke et al., 2020, Lively et al., 2021, Scotti et al., 2021). Understanding the mechanisms that give rise to these seemingly opposite biases has thus become a central focus in the field.

To explore the underlying mechanism of the two types of VWM bias, Chunharas et al. (2022) conducted a series of experiments and proposed an adaptive framework. This account posits that the two seemingly opposing biases (attraction and repulsion) are not random errors of the memory system. Instead, they reflect an adaptive strategy employed by the brain to minimize overall memory errors across different contexts. Specifically, when memory targets are similar, the nature of individual memory representations determines the resulting bias: if these representations are noisy and uncertain, the memory system shifts them toward more stable collective information (such as the ensemble average). This process enhances overall memory robustness and reduces average recall error, thereby resulting in attraction bias. In contrast, if individual memories are strong and precise, the memory system tolerates minor inaccuracies as a “cost” to obtain the “benefit” of avoiding catastrophic confusion between highly similar targets, which leads to the repulsion effect.

From the perspective of neural activity, dynamic neural field and bump-attractor models (e.g., Johnson et al., 2006, Johnson et al., 2009, Johnson et al., 2022, Secer et al., 2025, Wimmer et al., 2014) propose that when item features are highly similar, the local self-excitatory domains of their corresponding neural activity peaks or “bumps” overlap, creating a shared excitatory platform. This results in a systematic shift of peak locations toward the feature average, leading to fusion and behavioral signatures of memory attraction. In contrast, the repulsion effect, which also occurs between similar items, serves to maintain distinct neural representations. Functionally, this process acts to prevent their perceptual or mnemonic confusion. This is achieved through lateral inhibition (Carpenter and Blakemore, 1973, Rauber and Treue, 1998, Kiyonaga and Egner, 2016), whereby coexisting activity peaks exert mutual repulsive forces. These forces drive the peaks apart along an inhibitory gradient over time, enhancing representational separation and ultimately manifesting as memory repulsion.

Consequently, these models provide a unified view: the brain adapts its memory system to handle internal noise and prevent confusion, with this adaptation implemented by bump-attractor network dynamics in which noise-driven overlap causes attraction, while lateral inhibition leads to repulsion. It should be noted that these models implicitly presuppose a critical, yet often unstated, premise for generating a systematic repulsion bias: directional relationship in the feature space. This directional relationship refers to the memory system’s implicit encoding of the relational vector between items along a continuous feature dimension, such as whether a target is “redder” or “bluer” than a competitor. It is this directional relationship that guides the lateral inhibition process to yield a consistent repulsive shift. In the absence of such directional relationship, the motive to “avoid confusion” lacks the necessary spatial guidance and could lead to no memory bias or attractive errors, failing to achieve the functional segregation of memory traces. Of course, it should be emphasized that this directional relationship in the feature space can be extracted only when the target representations are clear. Conversely, when memory representations are noisy and uncertain, the system cannot establish the precise directional relationship. Instead, it prioritizes enhancing overall memory robustness and minimizing average recall errors, thereby resulting in attraction bias.

The aforementioned models and evidence provide a compelling account of how mnemonic interactions unfold among multiple task-relevant targets. However, a crucial ecological complexity remains: in real-world scenarios, targets are seldom encountered in isolation but are often embedded among task-irrelevant distractors that must be ignored. This reality raises a critical question: how do the adaptive mechanisms of attraction and repulsion, as described in these models, operate when the source of similarity is an irrelevant distractor rather than a fellow target? Research indicates that similar distractors can indeed distort the memory content of a target, inducing memory biases (Lorenc et al., 2018, Mallett et al., 2020, Rademaker et al., 2015, Wildegger et al., 2015). For instance, researchers sequentially presented a target and an irrelevant distractor, instructing participants to recall only the orientation of the initial target while ignoring the subsequent distractor. The results demonstrated that the target orientation was biased towards the distractor (Rademaker et al., 2015, Wildegger et al., 2015, Lorenc et al., 2018), a phenomenon known as memory attraction that has also been observed across various stimuli including color (Nemes et al., 2012, Saito et al., 2023), spatial location (Van der Stigchel et al., 2007), spatial frequency (Huang and Sekuler, 2010, Nemes et al., 2011), and even in faces (Mallett et al., 2020).

In these aforementioned studies, the target and distractor were presented sequentially, such that the target was no longer visible when the distractor appeared. When the target is stored in the maintenance stage, some researchers found that memory representations deteriorate over time (Rademaker et al., 2018, Shin et al., 2017). Moreover, the abrupt onset of a distractor can cause perceptual interference with the stored target representation (Bettencourt & Xu, 2016). This corruption can be understood as the introduction of noise or uncertainty into the target's memory representation (Lorenc et al., 2018). Critically, because the subsequently presented distractor produces a strong signal, the memory system, as proposed in the adaptive framework (Chunharas et al., 2022), adopts a strategic compromise: it biases the compromised and uncertain target representation toward the value of the salient and highly similar distractor to optimize overall memory performance, thereby producing an attraction effect. Thus, this attraction effect is further intensified when participants perceive a higher degree of similarity between the target and the distractor (Saito et al., 2023).

In the simultaneous presentation paradigm, however, target representations are less degraded by noise, thereby enabling the extraction of their directional relationship within the feature space. This extracted relationship can allow a distractor to serve as a spatial reference, leading the memory report of the target to be repelled away from it. This helps to minimize confusion between the target and the distractor, thereby plausibly generating a repulsion effect. Unfortunately, most previous studies investigating VWM filtering mechanisms have primarily focused on whether distractors impair memory performance for targets (Dube et al., 2017, Feldmann-Wüstefeld and Vogel, 2019, Jeong and Xu, 2013, Liesefeld et al., 2020, Matsuyoshi et al., 2010, Matsuyoshi et al., 2012). Notably, in these studies, the targets and distractors were dissimilar, and no findings regarding memory bias were reported. To our knowledge, only Golomb (2015) used a pre-cue to distinguish targets from distractors and reported a complex pattern of memory bias in the simultaneous paradigm. Specifically, four items were presented at two pre-cued locations, with two serving as targets and two as non-memorized distractors. Results revealed an attraction effect, where the recalled color of a target was biased toward a similar, non-memorized distractor. However, this also reflected a repulsion effect away from the other target, which was independent of the non-memorized distractors.

Based on existing studies, it remains unclear whether a distractor can serve as a reference point and induce a shift in the reported color of the target away from the distractor, potentially resulting in a repulsion effect. The current study investigated this issue. However, since distractors are typically required to be ignored, it is necessary to enhance their salience to prevent complete disregard, thereby facilitating access to the directional relationship in the feature space. To achieve this, the current study employed a simultaneous presentation paradigm and enhanced the salience of the distractors either by increasing the number of identical distractors or by altering their spatial configuration. This approach aimed to increase the likelihood of accessing the directional relationship in the feature space, thus allowing the identical distractors to function as a reference point. Specifically, Experiment 1 examined whether a repulsion effect occurs when memorizing a single color target and, crucially, whether this effect is modulated by the number of identical distractors. Experiment 2 investigated whether the number of identical distractors also influences memory bias when participants memorize four similar color targets. Experiment 3 manipulated the degree of spatial clustering among targets to rule out its potential influence on memory bias. Finally, Experiment 4 varied the spatial configuration between targets and distractors to further explore how the salience of distractors affects memory bias.