It has been proposed that atypical face scanning underlies poor recognition of facial identity in ASD 
but there are few direct tests of this hypothesis. In the current study, eye-movements of ASD and typically developing children were monitored while they learnt pictures of faces. In both groups, the best predictor of performance was the number of times participants’ eye-gaze moved into, and out of, core-feature interest areas (), indicating that successful face recognition is correlated with a pattern of multiple saccades between the core facial features.
We found no significant difference in the amount of time the groups spent looking at the eyes. This contrasts with reports of reduced looking time on the eyes when individuals with ASD watch video clips 
, although it is consistent with other studies of ASD using static facial images 
. However, the key question here was whether individual differences in looking time at the eyes was related to recognition performance, and we found no evidence to support this hypothesis in either ASD or typically developing children ().
The benefit of focusing on all the core features and not at the non-features converges with the most compelling result from our data – that greater movement between core features was highly associated with face recognition ability. Subsequent analyses showed that this result was not mediated by indices of general cognitive ability or degree of autistic symptoms. Moreover, although the number of participants in each group was relatively small, the significant association was clearly present in both ASD and typically developing groups, which demonstrates that the effect is replicable across samples. We suggest, therefore, that moving eye-gaze between the core features of a face is a crucial factor in face recognition in ASD and typically developing children.
Charawarska and Shic 
drew similar conclusions, suggesting that if an ASD child (age 2 or 4 years) focused exclusively on the eyes without distributing attention to all the core-features, their recognition ability would be compromised. However, we note that any causal relationships between movement of eye-gaze, and face recognition ability are currently undetermined. That is, aberrant scanning might lead to poor face recognition, or might be a consequence of an individual’s already poor face recognition skills. Alternatively the relationship may reflect some common factor underlying both reduced eye-movements and poor recognition ability.
Nevertheless, previous studies have suggested moving eye gaze between facial features allows spatial relations to be determined, and that a failure to do this inhibits the formation of a unified visual percept of a face 
. Such configural or holistic information is thought to be particularly important for accurately discriminating between facial identities 
and it is interesting to note that a number of studies have reported reduced holistic processing of faces in ASD. For example, it was found that ASD children were just as good at recognizing facial features presented in isolation as typically developing controls, but were worse than controls at recognizing features presented in the context of whole faces 
, suggesting they were less inclined to make use of the available information of spatial relations 
. Thus, in our participant sample, the apparent association between a lack of movement and poor face recognition skills might be explained by reduced use of configural/holistic face information.
An interesting comparison here is with individuals with developmental prosopagnosia (DP), a condition in which impaired face recognition occurs in the absence of any acquired brain damage, and in the context of normal low-level visual functioning 
. In a case study of adults with DP 
, aberrant patterns of face scanning were recorded, with attention being directed away from the internal configuration of core features, and towards peripheral face regions. The authors hypothesized that the disorganized, abnormal scan paths might underlie the impaired face recognition skills that characterize the condition. This hypothesis is largely consistent with results of our own study, however here we are able to demonstrate that it is dynamic saccades between core features that most strongly predict recognition ability, and also to extend the findings to at least two other participant populations. Progress in understanding this relationship will likely be made if experimental results across multiple developmental disorders are considered in combination with individuals developing typically.
The variable face recognition scores within the ASD group demonstrate that facial identity recognition difficulties, like virtually all ASD symptoms, are not present in all individuals on the autistic spectrum therefore its utility as a potential diagnostic marker of the condition is limited. However, poor face recognition skills are clearly evident in many ASD children, thus possible interventions are worth considering. In one study, Schmalzl, et al 
reported aberrant scan paths in a 4-year-old child initially identified as having DP, but who was later found to meet diagnostic criteria for ASD 
. An intervention program aimed at directing the child’s attention towards the core features of faces led to significant improvements in recognition of familiar and unfamiliar faces at follow-up assessment one month later 
. Whilst no comparison data was available, their findings suggest that a failure to attend to core features of the face was indeed a significant factor underlying poor recognition ability in this child.
As noted earlier, our experiment was designed to examine scan paths on faces during a learning phase, however studies such as 
suggest that scan paths during recognition
are also important. Furthermore, it is currently unclear how scan paths during encoding and recognition differ from each other, making this is an important question for future research.
In this study we directly tested hypothesized correlations between visual scan paths of faces, and recognition memory for faces in ASD and in typically developing children. Our analyses revealed that superior recognition performance was strongly associated with the degree of eye-movement between the core features of a face during encoding. Future research would be well placed in confirming the causal directionality of this association, which may then provide a useful basis for developing intervention techniques to effectively improve face recognition ability in ASD and other populations.