The research proposal aims to significantly advance our understanding of the core underlying deficits observed in Autistic Spectrum Disorder based on a novel theoretical model of autism, and using non-invasive eye-tracking technology to obtain more rich and exact data. The theoretical proposal is twofold, on the one hand we hypothesise that autism may be characterised by biologically based dysfunction of mechanism responsible for identifying aspects of the external world that respond contingently with one's own behaviour. In normal development, a shift in contingency preference - from perfect to high but imperfect response-stimulus contingency - occurs at around three months of age. We hypothesise that in ASD this preference shift does not take place, leaving individuals with ASD blind to, or with decreased sensitivity to, imperfect contingency in relation to their behaviour. On the other hand, contrary to the mainstream view, we propose that the deficits ASD individuals have in understanding social partners as intentional agent guided by unobservable mental states might not be due to a basic deficit in computing the representational states of others. Instead, they might have problems in using the automatically computed belief-representations for future social interactions. The proposed methodology and studies would lead to earlier diagnosis and highly targeted intervention programs for ASD, and therefore to improved outcomes for children and families.