Background: The Picture Exchange Communication System (PECS) is a common treatment choice for non-verbal children with autism. However, little empirical evidence is available on the usefulness of PECS in treating social-communication impairments in autism. Aims: To test the effects of PECS on social-communicative skills in children with autism, concurrently taking into account standardized psychometric data, standardized functional assessment of adaptive behaviour, and information on social-communicative variables coded in an unstructured setting. Methods & Procedures: Eighteen preschool children (mean age = 38.78 months) were assigned to two intervention approaches, i.e. PECS and Conventional Language Therapy (CLT). Both PECS (Phases I-IV) and CLT were delivered three times per week, in 30-min sessions, for 6 months. Outcome measures were the following: Autism Diagnostic Observation Schedule (ADOS) domain scores for Communication and Reciprocal Social Interaction; Language and Personal-Social subscales of the Griffiths' Mental Developmental Scales (GMDS); Communication and Social Abilities domains of the Vineland Adaptive Behavior Scales (VABS); and several social-communicative variables coded in an unstructured setting. Outcomes & Results: Results demonstrated that the two groups did not differ at Time 1 (pre-treatment assessment), whereas at Time 2 (post-test) the PECS group showed a significant improvement with respect to the CLT group on the VABS social domain score and on almost all the social-communicative abilities coded in the unstructured setting (i.e. joint attention, request, initiation, cooperative play, but not eye contact). Conclusions & Implications: These findings showed that PECS intervention (Phases I-IV) can improve social-communicative skills in children with autism. This improvement is especially evident in standardized measures of adaptive behaviour and measures derived from the observation of children in an unstructured setting. (Contains 5 tables.)