“All spiders are equally scary, but some spiders are more equal than others.” This paraphrase from Orwell’s Animal Farm may represent the summary of a few studies that investigated the features of spiders that make them fearful (Davey, 1991; Lindner et al., 2019). Although any spider can somehow elicit a phobic reaction in spider-fearful people, participants in some studies have referred to being specifically more sensitive to some perceptual features (in particular, the locomotion pattern; Davey, 1991; Lindner et al., 2019) than to others. These self-reported preferences found indirect confirmation in behavioral studies describing attentional or interpretational biases for spiders depending on their movement trajectory, either actual (Basanovic et al., 2017) or perceived (Riskind et al., 1995).

The features that make spiders fearful and/or disgusting have been proposed to be cross-cultural characteristics (Davey et al., 1998) that distinguish them from similar animals (Landová et al., 2021), including from those sharing comparable dimensions and poisonousness (e.g., bees and wasps, in Gerdes et al., 2009); however, Frynta et al. (2021) suggested that arachnophobia could derive from a more evolutionarily-relevant fear for scorpions.

Interestingly, these preferences do not seem to have any rational ground (e.g., they are not specific characteristics of poisonous spiders), coherently with the description of phobia adopted in DSM 5: a disorder characterized by an exaggerated fear of a certain stimulus or animal (American Psychiatric Association, 2013). Despite the lack of rational links between perceptual features of spiders and their dangerousness, arachnophobia is ranked as one of the most common specific phobias (Eaton et al., 2018). The higher prevalence of phobias related to evolutionary-relevant stimuli (e.g., spiders and snakes) than to culturally relevant stimuli (e.g., atomic bombs and guns) supports the idea of an innate preparedness for phobias (Seligman, 1971).

The idea that specific phobias are hardwired in the nervous system (LeDoux, 1994) fits with the evidence that brain circuits related to the categorization of spiders differ from those related to the categorization of snakes (Taschereau-Dumouchel et al., 2018). The existence of a brain circuitry decoding for specific animals implies that the perceptual features mostly characterizing evolutionary-relevant stimuli could trigger a stronger fear response than generic features shared with evolutionary-irrelevant stimuli. This could happen at an unconscious level, with relevant clinical implications for the optimization of exposure therapies (Taschereau-Dumouchel et al., 2018; Baroni et al., 2021; Frumento et al., 2021, 2022; Cesari et al., 2023). When facing a spider, emotions and the consequent behaviors could be elicited with little or no role of cognitive, conscious processes. Deepening the role of consciousness in coupling emotions and behaviors could help understand specific phobias and improve therapeutic approaches.

In line with this scope, it has been recently proposed that investigating the perceptual features that make spiders scary (Frynta et al., 2021) and checking for their possible association with phobic symptoms (Lindner et al., 2019) could shed light on the evolutionary origins of spider phobia. This association appears to be non-linear, suggesting a ceiling effect (Lindner et al., 2019); most participants rated the movement pattern as the perceptual feature that is most responsible for the perceived dangerousness of spiders (Lindner et al., 2019). However, the authors recognize that these results were limited to self-reports related to imaginary spiders, for which only one feature (i.e., locomotion) was investigated.

To overcome the limitations affecting all studies that investigated differences in scariness due to perceptual features of phobic stimuli, an ecological experimental setting was developed in augmented reality, allowing participants to manipulate various features of hyper-realistic virtual spiders in order to make them look dangerous or harmless. Primarily, volunteers were characterized by different levels of fear of spiders, and their preferences were analyzed taking into account the lack of linear association with phobic symptoms previously reported (Lindner et al., 2019).

Among the 56 volunteers initially recruited, 10 were excluded based on the clinical threshold of psychopathological symptoms different from specific phobias revealed by the SCL-90-R. One participant was further excluded for suspending the ongoing session due to an unsustainable aversiveness for the experimental setup. The final sample considered for analyses was then reduced to 45 participants (12 men, 33 women; mean age = 26.2 years, SD = 5.5), with a level of arachnophobia symptoms—as measured through the SPQ—continuously spread from 0/30 to 26/30. The higher rate of females (73.3% of the final sample) is consistent with the higher prevalence of specific phobias in women (Eaton et al., 2018).

Table 1 summarizes the perceptual preferences expressed by all participants about dangerous and harmless spiders, comparing the three statistics (mean, median, and MMPD function) and the incoherency value described in section 2.5. Spiders modified to be as fearful as possible (red-filled cells in Table 1) were hairy, thick, and moving in a spider-like pattern. Particularly, the average preferences expressed by participants to make spiders as harmless as possible (green-filled cells in Table 1) resulted in spiders that were hairless, slim, and moving in a butterfly-like pattern. Incoherency between the preferences expressed for harmless and dangerous spiders was low for all features.

The correlation matrix of the three features (the hairiness, the body/leg size, and the locomotion pattern, each in both its dangerous and harmless variations) and the fear of spiders revealed the existence of significant relations (Figure 2). In particular, the preferences expressed for each feature when making it dangerous or harmless resulted in being significantly and negatively correlated; moreover, a positive correlation between the level of fear of spiders (SPQ) reached (p < 0.001) and approached (p = 0.05) significance with the preferences expressed about the hairiness and the body/leg size of dangerous spiders; the preferences expressed for the hairiness and body/leg size of both the dangerous and the harmless spiders were significantly correlated (see Figure 2).

Figure 3 shows individual data (each dot corresponds to a subject), previously summarized in Table 1, and models them with a normal heteroskedastic spline-based model to (1) deepen the relations between each perceptual preference and the level of fear of spiders (SPQ score) and (2) further integrate this information with the results of a hierarchical cluster analysis. In fact, for each row of the figure, two plots describe how each feature was modified to create dangerous (Figures 3A,D,G) or harmless (Figures 3B,E,H) spiders. A third plot describes the distribution of inter-individual incoherency in perceptual preferences in relation to the level of spider fear (Figures 3C,F,I). Finally, each plot indicates with different colors of the dots, the presence of multiple groups resulting from cluster analysis.

Integrating the information conveyed by Figures 2, 3 shows that the preferences expressed about the hairiness and the body/leg size of dangerous spiders are very similar among participants with low SPQ scoring, compared to those with high SPQ scoring (Figures 3A,D). The cluster analysis based on each feature (i.e., the hairiness, the body/leg size, and the locomotion pattern) and each instruction (i.e., making the spider as dangerous or as harmless as possible) revealed that in most cases (Figures 3B,D,E,G,H), two distinct groups were characterized by preferences polarized on the two possible extremes. Finally, the inter-individual incoherency between hairiness preferences related to dangerous and harmless spiders showed that the participants with higher levels of fear of spiders were less coherent than those with low or null levels of fear of spiders (Figure 3C). On the other hand, the incoherency between preferences concerning the body/leg size and locomotion was not affected by the fear of spiders (Figures 3F,I).

It is worth noting that Figure 3 represents the preferences calculated through the MMPD function described in the “Data analysis and representation” section; however, preferences summarized through means (Supplementary Figure S3) or medians (Supplementary Figure S4) show a comparable distribution. Figure 3 uses different colors to represent a clusterization based on one factor per plot, whereas the results related to other clustering criteria are reported in Supplementary materials. Supplementary Figures S5–S7 show the same data of Figure 3 clustered for all factors (the three perceptual features and the SPQ score), all perceptual features, and the harmless/dangerous version of each feature, respectively.

Finally, the orientation of spiders (frontal, lateral, or posterior) resulted in slightly different preferences about the hairiness of dangerous spiders, which was on average lower for spiders approaching the participant than for those sliding or withdrawing: preferences about the body/leg size and the locomotion were almost overlapped regardless of orientation (Supplementary Figures S1, S2).

The present study reports behavioral evidence that specific preferences in perceptual features can make a virtual spider look more or less dangerous/harmless. Importantly, some of these preferences are correlated with the level of fear of spiders.

On average (Table 1), fearful spiders were hairy, thick, and moving according to a spidery pattern (Figure 3L); particularly, harmless spiders were hairless, slim, and moving according to a butterfly-like pattern (Figure 3M). However, the relationship between phobic symptoms and sensitivity to each perceptual feature is more nuanced. In particular, Figure 2 reports significant correlations between many couples of variables, the most interesting of which showed that (1) less spider-fearful participants found hairy spiders to be more dangerous, and (2) spiders modified to look dangerous were both thick and hairy, despite most of the other features are mutually independent.

In fact, the unrecognizable grouping resulting from the different clustering criteria shown in Supplementary Figure S7 suggests that the preferences shown for each feature are mutually independent (i.e., dangerous spiders can be slim or thick regardless of their hairiness). To the contrary, Figures 2, 3C,F,I (in particular, significant correlations between each feature’s harmless and dangerous preferences) show that each perceptual feature was characterized by a high internal coherency (i.e., participants who made harmless spiders slim also made fearful ones thick): hairiness’ coherency lowered at the increase of fear of spiders (Figure 3C), thus exemplifying the generalization effect that characterizes phobic patients according to DSM 5 (American Psychiatric Association, 2013).

When clustering criteria were limited to each specific feature (e.g., the hairiness of dangerous spiders), two groups of participants could be recognized as expressing preferences polarized on the two possible extremes (e.g., Figure 3B), corroborating the lack of a linear relationship between arachnophobia symptoms and features’ preferences previously reported by Lindner et al. (2019). The group membership was in most cases not related to the level of fear of spiders; however, for both the body/leg size of dangerous spiders (Figure 3D) and the hairiness of harmless spiders (Figure 3B), the minority group was entirely composed of the participants with mild or severe (SPQ score ranging between 10 and 30) phobic symptoms.

In contrast with this polarization of preferences, the hairiness was the feature revealing the smallest differences among clusters (Figure 3A) and the higher incoherency in answers (Figure 3C): these results—and, to a lesser extent, those related to the body/leg size (Figures 3D,F)—can be traced back to a generalization effect increasing at the increase of phobic symptoms (as expected from phobic patients; American Psychiatric Association, 2013) along a continuum.

The preferences expressed about the locomotion patterns (Figures 3G,H) are very different from those expressed for both hairiness and body/leg size. Indeed, in line with previous studies (Davey, 1991; Lindner et al., 2019), the vast majority of participants—regardless of their level of fear of spiders—considered spiders as dangerous when moving according to a spider-like pattern. In addition to this, the different perceptual preferences characterizing spiders shown in a frontal orientation from those shown in a lateral or posterior orientation (Supplementary Figures S1, S2) support previous findings showing that spiders crawling toward the participant elicit reactions different from the withdrawing ones (Lindner et al., 2019). In this regard, the feature most heavily affected by a spider’s orientation was dangerous hairiness, whereas preferences concerning the body/leg size and the locomotion almost overlapped (Supplementary Figures S1, S2).

Finally, the possibility of modeling flying spiders—which should have always been considered harmless to coherently mirror dangerous ones— has confused participants, resulting in the highest inter-individual incoherence among the three features (Figure 3I). This inter-individual incoherency in preferences could be explained by the locomotion pattern which is the only perceptual feature that could break realism since it ranged from a spider-like pattern to an impossible butterfly-like pattern.

As a final consideration, while these results are supported by the strengths coming from the introduction of an objective behavioral assessment based on augmented reality (more objective than the self-report tools used so far in the scientific literature about phobic perceptual preferences), the whole study inevitably comes with some noteworthy limitations that impose caution in generalizing its results. These limitations—better detailed in the next section—will be overcome only by a wider replication of the study, made possible by the full sharing of its complete data.

The present study is the first one attempting to objectively assess the preferences in perceptual features that make a prototypical phobic stimulus (a spider) more or less fearful and to link these preferences to arachnophobia symptoms. This topic has been so far investigated mostly through subjective self-reports, which are not as related as commonly thought to behavioral and physiological parameters (Baldini et al., 2022; Frumento et al., 2022) and could fail to address complex relationships between variables. With respect to the previous scientific literature, the use of augmented reality made it possible to change the perceptual features of spiders in real time, thus reaching the best compromise between ecological validity and experimental manipulation (Iannizzotto et al., 2024). In fact, the holographic tridimensional spiders were highly realistic, as confirmed by the aversiveness of the experiment reported by the most spider-fearful participants (one of whom felt the need to abandon the experimental session after the first trial).

On average, spiders modified to be dangerous were hairy, thick, and moving following spider-like locomotion (Figure 3L), whereas spiders modified to be harmless showed opposite features (Figure 3M). However, these preferences underlie more complex relationships between each perceptual feature and the level of fear of spiders. In particular, inter-individual incoherency in hairiness preferences was increasing with the increase of fear of spiders, consistent with the generalization tendency described in DSM 5 (American Psychiatric Association, 2013). In addition, the locomotion of dangerous spiders showed a ceiling effect consistent with previous evidence showing that spider-like movements are the most feared, regardless of phobia level (Lindner et al., 2019). This suggests that the generalization effect is only valid for realistic features and thus does not apply to spiders with a butterfly-like locomotion pattern.

Considering that flying spiders do not exist in nature, this last result assumes even more relevance. In fact, a rationality-driven decision process would consider a flying spider to be more dangerous than a non-flying one—the more the abilities, the more the potential harm. Nevertheless, holographic spiders induced a higher fear when they were mostly resembling the unique perceptual features of a prototypical spider that is clearly distinguishable from all other insects for its hairiness, movement pattern, and thick body/leg proportions.

These results suggest that conscious cognitive processes and rational thinking are not driving the preferences concerning which perceptual features make a spider dangerous or harmless. In fact, both spider-fearful and non-spider-fearful participants considered spiders dangerous when their appearance matched that of an ancestral menace (e.g., crawling in a spider-like pattern). In addition to that, the group of spider-fearful participants felt threatened by a wider range of spidery features (e.g., low hairiness or slim body/leg size) that do not seem to rely on any rational ground (e.g., none of these features is a reliable marker of spider poisonousness or aggressiveness).

These unconsciously driven preferences support the idea that spider phobia stems from innate mechanisms aimed at quick and dirty detection of evolutionary-relevant animals (LeDoux, 1994). However, the present study is not designed to unambiguously discriminate whether these preferences are culturally or biologically driven. Future studies implementing a subliminal administration of spider-like stimuli differing for perceptual features could provide further information about the nature of these preferences by getting around to conscious processing—most likely affected by cultural biases.

Finally, the recent frontier of generative artificial intelligence opens up the possibility of using the pictures produced for each trial (made publicly available together with the corresponding raw data) to train systems aimed at creating phobic stimuli that elicit different intensities of fear to improve the acceptability of exposure therapies and the generalizability of their therapeutic outcomes.

The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found below: The Unity software for the “spider-modifier” used in the experiment is publicly available on GitHub repository (https://github.com/M45K/RitualOfChud). Data and analysis scripts (performed in R software) are publicly available at the Open Science Framework repository (https://osf.io/73s2q/?view_only=c9b217b1ad3b433faa271a496cc08f71).

The studies involving humans were approved by Comitato Bioetico dell’Università di Pisa. The studies were conducted in accordance with the local legislation and institutional requirements. The participants provided their written informed consent to participate in this study.

SF: Conceptualization, Data curation, Formal analysis, Writing – original draft. PF: Data curation, Formal analysis, Methodology, Writing – review & editing. ML: Supervision, Writing – review & editing. DM: Supervision, Writing – review & editing, Data curation, Methodology. AG: Conceptualization, Funding acquisition, Project administration, Resources, Supervision, Writing – review & editing.