Past research has shown that visual stimuli capture attention based on factors such as task relevance (Folk et al., 1992; Büsel et al., 2020), visual salience (Nothdurft, 1993; Itti et al., 1998; Wang and Theeuwes, 2020), or prior individual experience with these stimuli (Awh et al., 2012). It is more contested, however, if some visual stimuli can capture attention based on their general phylogenetic relevance (cf. Öhman, 1993; Öhman et al., 2001), a prime example of which is the capture and holding of attention by specific emotional facial expressions (e.g., Eastwood et al., 2001; Fox et al., 2001; Vuilleumier, 2005; Bannerman et al., 2010; Van Hooff et al., 2013). Critically, several studies suggested that fearful faces could capture attention even if task-irrelevant (Bannerman et al., 2010) and that this effect could not be ascribed to visual salience alone (Khalid et al., 2017). Khalid et al. (2017) carefully controlled for mere influences of salience by presenting faces as cues either in their cardinal orientation or upside-down. The authors found that task-irrelevant fearful faces only captured attention if presented in their cardinal orientation, indicating that emotion recognition was relevant for the capture of attention (cf. Huynh and Balas, 2014). Salience, in contrast, was insufficient to explain the capture of attention, as salience was the same for cardinal and inverted orientations. The observation of salience independence is critical, as otherwise the visual salience of a particular facial expression itself could capture attention merely through basic visual and emotion-unspecific characteristics (cf. Morris et al., 2002; Horstmann and Bauland, 2006).

Apart from task-irrelevant capture by liminal emotional faces, researchers have looked into electrophysiological data during task-relevant perception of subliminally presented faces at central locations. Findings for early components were indecisive, with some studies showing early differential processing of fearful compared to neutral facial expressions (Kiss and Eimer, 2008; Pegna et al., 2008; Smith, 2012; Zhang et al., 2012, 2017) while others do not (Liddell et al., 2004; Walentowska and Wronka, 2012). In addition, subliminal processing was only verified on a trial-by-trial basis in some studies (Pegna et al., 2008; Smith, 2012; Zhang et al., 2012), of which two found an increased N170 for subliminal fearful faces (Pegna et al., 2008; Smith, 2012).

Here, we went one step further and tested the important question of whether the known capture effects of task-irrelevant fearful faces observed by Khalid et al. (2017) extend to subliminally presented emotional faces, as the general human ability to differentially process subliminal emotional expressions would suggest (cf. Kiss and Eimer, 2008; Pegna et al., 2008). The question is important considering alternative theories arguing for prominent roles of awareness independence vs. awareness dependence in human social and emotional processing abilities. On the one hand, some theories argue that the rapid processing of emotional facial expressions provided a phylogenetic benefit pre-dating consciousness and, thus, the corresponding capture of attention by emotionally significant facial displays could be awareness-independent (Morris et al., 1998; De Gelder et al., 1999; Öhman, 2002; Dolan and Vuilleumier, 2003; Tamietto and De Gelder, 2010). On the other hand, some theories emphasized the role of awareness or consciousness in some social and emotional skills (Frith and Frith, 2007; Frith, 2008). Authors argued for a role of awareness at least for forms of more deliberate social processing, taking intentions into account. Likewise, emotions themselves have been described as the result of both, more automatic and more reflective processes (see, for example, Weiner, 1986). The higher cognitive processes are involved when we have to make sense of the behavior of others. Now, it might seem that the processes triggered by stimuli, such as specific emotional faces, could be relatively simple, sparing the more complex forms of emotional and social processing that depend on awareness. However, in this case it is also important to concede that manipulations of awareness and visibility carry the risk of sacrificing much of this processing efficiency (cf. Hedger et al., 2016).

In the present study, we therefore tested whether masked (subliminal) fearful faces capture attention. In contrast to the classic dot-probe task, where emotional faces are used as cues and an onset stimulus is shown as target (e.g., a dot as location target or an orientated shape as discrimination target; see Mogg and Bradley, 1999; Holmes et al., 2005; Cooper and Langton, 2006; Brosch et al., 2008; Puls and Rothermund, 2018), we used two stimuli in the target display, instead of only one onset target (cf. Wirth and Wentura, 2018, 2019). We applied this variation of the dot-probe paradigm, a visual search task for a predefined face target presented unforeseeably at one of two positions, following Khalid et al. (2017). The visual search for target faces has some decisive advantages compared to the classic dot-probe task with onset targets. In particular, the task of searching for a target face allowed us to vary the task relevance of the masked (subliminal) fearful faces, as we will explain in more detail below.

To start with, prior to the face target, we presented a fearful face as a cue either at the position of the incumbent target (valid condition) or at the alternative position (invalid condition). Although the cues were on average not predictive of the target position, in line with prior studies with supraliminal fearful face cues (cf. Khalid et al., 2017), we expected a validity effect if the subliminal fearful face cues captured attention, too: faster responses to validly cued targets than to invalidly cued targets. However, we were aware that in the small cue-target interval more complex processes could occur than a mere capture of attention and dwelling of attention at the cued position lasting until (or even after) the target is presented. For example, it is possible that cues initially capture attention, but that attention is then deallocated from the cue and shifted back to a neutral position prior to the target (Theeuwes et al., 2000). Likewise, it is possible that the fearful face cue captures attention but that target search is then so easy that one would literally not see any evidence of attention dwelling at the cued position, once target displays have commenced (Gaspelin et al., 2016). Therefore, for our test of attention capture by subliminal fearful face cues, we used a more exhaustive measure than mere reaction time effects. We used the electroencephalogram (EEG) to test for attention-elicited event-related lateralizations (ERLs; Luck and Hillyard, 1994; Luck, 2012). For example, Eimer and Kiss (2007) found attention capture by irrelevant fearful faces in the form of an N2pc, a stronger negativity at about 200–250 ms following stimulus onset at contra-compared to ipsilateral posterior scalp sites that was reflective of an attention shift to stimuli presented to the left or to the right. In addition, such N2pc effects can also be elicited by subliminal stimuli (e.g., color cues; Ansorge et al., 2009). To find out whether a subliminally presented fearful face cue captures attention, we conducted two cueing experiments (Posner, 1980) and used EEG to look for cue-elicited ERLs. In Experiment 1, we used task-irrelevant fearful and disgusted faces as cues, each presented with a task-relevant neutral face at the opposite position. In Experiment 1, we would have expected validity effects by masked fearful faces and maybe by masked disgusted faces if these subliminal cues captured attention in a similar way as their supraliminal counterparts used in Khalid et al. (2017). These authors found orientation-specific cuing effects for fearful faces but orientation-independent cueing effects for disgusted faces, showing that capture by fearful faces is emotion-specific. In the present Experiment 1, participants had to search for neutral face targets, meaning that the fearful face cues were task-irrelevant. However, we admit that we thereby created a relatively conservative test of the fearful face cues’ potential to capture attention, as the task-relevant neutral face stimulus on the opposite side of the fearful face cue would have matched the top-down attentional control settings (or search criterion) of the participants for the targets. Thus, if neutral face stimuli presented opposite to the fearful face cues would capture attention in a top-down contingent way, this would counteract the capture of attention by the task-irrelevant face cue. Therefore, in Experiment 2, we made the negative facial expression of fear task-relevant, in order to boost attention-capture effects by subliminally presented fearful face cues. In both experiments, face cues were presented in upright or in inverted orientation. As in Khalid et al. (2017), this was done to rule out salience as a potential explanation of any cueing effect: If a capture effect was due to salience, we expected to find it both in upright and inverted conditions. However, a capture effect that was due to the specific emotional expression of the cues was expected to only show in the upright orientation, but not in the inverted orientation.

In Experiment 1, we wanted to find out whether task-irrelevant, subliminally presented fearful facial expressions automatically capture attention and to what extent this effect could be specific to fear vs. disgust. To note, fear has an arguably higher potential to capture attention in an emotion-specific (orientation-dependent) way (cf. Khalid et al., 2017). Participants were instructed to search for a neutral face. Prior to the target faces, we presented participants both neutral cue distractors and negative emotional (fearful or disgusted) face cues, which were oriented either upright or inverted.

In Experiment 1, neither the task-irrelevant masked negative emotional cues nor the search-goal congruent, task-relevant emotionally neutral face cues were strong enough to elicit attention capture effects. As masked emotional and task-relevant faces were presented simultaneously in the cueing display, their respective capture effects might have cancelled each other out. To boost the effect of the subliminal negative expressions, in Experiment 2, we changed the task: Now, participants had to search for a fearful expression in the target display (and report the orientation of a letter T there). In this situation, the masked fearful cue face became task-relevant, as it shared its type of emotional expression with the target. As the fearful cue face was both task-relevant and (evolutionary) salient, its ability to capture attention should have been enhanced. By the same logic, the masked neutral face in the cueing display was now task-irrelevant. It was expected to lose the power it may have had to capture attention in a top-down contingent way in Experiment 1. All in all, the chances for capture of attention by the subliminal fearful cue faces relative to the neutral distractor faces was, thus, enhanced in Experiment 2 relative to that of the fearful face distractors in Experiment 1.

To account for the relatively small sample size, we analyzed the N2pc component to the masked cue face pooled over Experiments 1 and 2. An ANOVA with the mean amplitude values from the separate analyses and the independent variables Hemisphere (contra- or ipsilateral) and Orientation (upright or inverted), again, revealed no significant effects, all Fs(1, 27) < 1.73, all ps > 0.200, all η p 2 s < 0.06. In addition, the BF₁₀ = 0.206 (main effect N2pc) yields substantial evidence for the H₀ (see Table 2 for a complete list of BF₁₀s). A post hoc power analysis for the N2pc shows that with an N = 28 (α = 0.05, power = 0.8, one-sided), medium-sized effects (down to 0.48) could have been found.

In addition, we combined the behavioral data from Experiments 1 and 2 and did pooled analyses with RTs and ACCs and the independent variables Validity (valid/invalid) and Orientation (upright/inverted), as a post hoc power analysis (α = 0.05, power = 0.8) for the separate analyses revealed that only big effects (Exp. 1: d = 0.73; Exp. 2: d = 0.65) could have been found. In the pooled analysis (N = 29), still no effects were found for the RTs, all Fs(1, 28) < 0.79, all ps > 0.383, all η p 2 s < 0.03. For the ACCs, in line with the separate results from Exp. 2, only the main effect of Validity was significant, F(1, 28) = 7.92, p = 0.006, η p 2  = 0.24 (valid: 76.8%; invalid 77.9%). See Table 4 for Bayes factors. Participants made more errors in valid than invalid cases. This might be caused by the relatively long interval between cue and target (350 ms), which was needed to analyze ERPs to the cue without distortion by the target screen. According to Klein (2000), this time interval might already subject processing at cued locations to some counteracting influences of inhibition of return.

To examine attention capture by subliminally presented fearful facial expressions, we conducted two experiments and analyzed electrophysiological (lateralized ERPs) and behavioral data (RTs and ACCs). In Experiment 1, we found no evidence for attention capture by task-irrelevant, subliminally presented fearful facial expressions, neither in ERPs, nor in behavioral data. As the ipsilateral activity after cue onset was numerically even more negative than the contralateral (see Figure 3), which would represent an N2pc to the task-relevant neutral face, the attention-capture effects by the task-relevant neutral and task-irrelevant negative faces might have diminished each other. Therefore, in Experiment 2, we made the negative facial expression (fearful face) task-relevant. If in Experiment 1, both task-irrelevant fearful and task-relevant neutral faces competed for attention, in Experiment 2, the effects of task relevance and fearful emotion should add up. Against our expectations, we again found no significant validity effects in Experiment 2. As even task relevance could not enable subliminal faces to capture attention, we can conclude that conscious perception is a prerequisite for attention capture by fearful expressions.

In general, some “pre-attentive” discrimination of visual features related to emotional facial expressions is possible for faces presented in the periphery (e.g., Pizzagalli et al., 1999; Rigoulot et al., 2011, 2012; Calvo et al., 2014; Towler and Eimer, 2015). Yet, the peripheral presentation of our faces might have impeded effects in later occurring ERPs, as peripheral presentation of faces at both peripheral locations somehow masks the strongly delayed ipsi- vs. contralateral ERPs (see Takamiya et al., 2020; Schindler et al., 2022). Also for centrally presented well visible fearful faces (presented for 50–100 ms; unmasked), differential effects disappear at the level of the N2/EPN when attending to overlaid perceptual information (Schindler et al., 2020a; Steinweg et al., 2021) or under conditions of peripheral load (e.g., see Schindler et al., 2020b, 2021).

At face value, the conclusion that awareness might be necessary for capture of attention by peripheral fearful faces cues (or other, e.g., neutral facial expressions if task-relevant) seems to be inconsistent with some past findings (Morris et al., 1998; Dolan and Vuilleumier, 2003; Tamietto and De Gelder, 2010; Rohr et al., 2012; Rohr and Wentura, 2021). However, it should be noted that many of these studies did not control for salience differences between different masked stimuli, and when this possibility was tested, it turned out that salience was probably involved in the processing of subliminal fearful faces (Morris et al., 2002). In contrast, in the current study, we took great care to equate different stimuli in terms of their salience. In addition, studies showing assumedly subliminal processing of fearful faces are often not using the most convincing visibility tests, so that some of the processing of these stimuli could be due to residual visibility of the fearful faces (cf. Hedger et al., 2016). Finally, we presented subliminal face cues at unattended positions to test whether they capture attention. In contrast, many prior studies with masked fearful faces presented the subliminal stimuli at attended locations (e.g., Rohr et al., 2012). To the degree that directing spatial attention to a face facilitates the processing of its features (but see Finkbeiner and Palermo, 2009, for evidence to the opposite), our procedure would simply not be sensitive to these kinds of attention-dependent processing.

The present results are, thus, also in line with a view that stresses that some forms of processing facial emotional expressions of humans could require awareness (Hedger et al., 2016). We note, however, that our procedure was challenging for the participants, as we always presented two faces side by side and, thus, two relatively interesting and informative stimuli competed for our participants’ attention. It could be that the chances for the expression of awareness-independent capture of attention by subliminal fearful faces are better under less challenging conditions. Yet, we also wanted to point out that supraliminal face cues would pass the test even with the currently employed relatively high demands (cf. Khalid et al., 2017).

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

Ethical review and approval was not required for the study on human participants in accordance with the local legislation and institutional requirements. The patients/participants provided their written informed consent to participate in this study.

MK and UA designed the experiments. TD programed the experiments. MK collected the data. DB analyzed the data. DB and UA wrote the manuscript. All authors contributed to the article and approved the submitted version.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.