Background: Studies of chromosome 15 abnormality have implicated over-expression of paternally imprinted genes in the 15q11-13 region in the aetiology of autism. To test this hypothesis we compared individuals with Prader-Willi syndrome (PWS) due to uniparental disomy (UPD--where paternally imprinted genes are over-expressed) to individuals with the 15q11-13 deletion form of the syndrome (where paternally imprinted genes are not over-expressed). We also tested reports that PWS cases due to the larger type I (TI) form of deletion show differences to cases with the smaller type II (TII) deletion. Method: Ninety-six individuals with PWS were recruited from genetic centres and the PWS association. Forty-nine individuals were confirmed as having maternal UPD of chromosome 15 and were age and sex matched to 47 individuals with a deletion involving 15q11-13 (32 had the shorter (T II) deletion, and 14 had the longer (TI) deletion). Behavioural assessments were carried out blind to genetic status, using the Autism Diagnostic Observation Schedule (ADOS), the Autism Diagnostic Interview (ADI), the Autism Screening Questionnaire (ASQ), the Children's Yale-Brown Obsessive-Compulsive Scale (CY-BOCS), the Vineland Adaptive Behaviour Scales (VABS), and measurements of intellectual ability, including the Wechsler and Mullen Scales and Raven's Matrices. Results: UPD cases exhibited significantly more autistic-like impairments in reciprocal social interaction on questionnaire, interview and standardised observational measures. Comparison of TI and TII deletion cases revealed few differences, but ability levels tended to be lower in the TI deletion cases. Conclusions: Findings from a large study comparing deletion and UPD forms of Prader-Willi syndrome were consistent with other evidence in indicating that paternally imprinted genes in the 15q11-13 region constitute a genetic risk factor for aspects of autistic symptomatology. These genes may therefore play a role in the aetiology of autism. By contrast with another report, there was no clear-cut relationship between the size of the deletion and the form of cognitive and behavioural phenotype.