Current diagnostic criteria for autism are guided by the 'Triad of Impairments' (Wing & Gould, 1979), of which 'social impairment' forms an individual category. Social difficulties have been observed in children with autism under the age of two years and the difficulties continue throughout adulthood. Few studies have used computer-based training packages (CBTP's) to facilitate social competency in individuals with autistic spectrum disorder (ASD). The present study was completed in two phases: Phase I -compared 14 individuals with ASD (IQ ý: 70), ten individuals with no disabilities (ND) and eight individuals with mild mental retardation (MR; IQ 50-70) on their ability to identify appropriate and inappropriate social interactions using 12 scenarios on a tailor-made DVD. Phase H- the ASD group was divided into two groups. Eight individuals were trained using six of the scenarios from Phase I and eight new scenarios. Following training, both groups with ASD re-viewed the 12 scenarios shown in Phase I. The results showed that individuals with ASD were less able_to identify appropriate and inappropriate social skills than individuals with MR (t (18) = 4.9, p<00 1) or ND (t (13) = 7.5, p<00 1). The severity of the autistic symptoms negatively impacted on performance (r (14) =- . 83, p<. 001). The training group significantly improved their performance (z = -2.52, p= . 05) on both trained (z = -2.207, p= . 027) and untrained scenarios (z = -2.032, p . 042). The untrained group scores showed no significant difference between Phase I and Phase II. The current study identified that individuals with autism have more difficulty understanding inappropriate and appropriate behaviour than their MR and ND peers. However the DVD was an effective method of training a component of social competency for this population. Within the trained group, ability improved on untrained as well as trained scenarios, this suggests a generalisation of skills rather than rote learning or practice effects. Future studies would benefit from addressing how to generalise this skill from computers to 'real-life' situations