Autism spectrum disorder (ASD) is a neurodevelopmental disorder with substantial phenotypic and etiological heterogeneity. 10-20% of cases are thought to be due to copy number variations (CNVs). To further identify the causal genes within these CNVs we developed a novel statistical framework to identify genomic regions that are recurrently deleted or duplicated in ASD. Analysis of CNV data from 19,663 autistic patients and 6,479 normal controls, identified 57 recurrently deleted and 62 recurrently duplicated regions associated with ASD, some of which show gender specific biases. Examining the phenotypes of patients with these 119 deletions and duplications in an independent cohort of patients with CNVs (DECIPHER) confirmed these regions are associated with ASD phenotypes, however also revealed distinct differences in the range of observed phenotypes for different duplications and deletions. To identify likely candidate genes within the significant genomic regions, we used the FANTOM5 expression atlas to prioritize those with enriched expression in brain. Of 830 coding genes found in the recurrently deleted regions and 518 in the duplicated regions, the expression of 37.3% and 37.4% are significantly enriched in brain. This is a significantly higher proportion than the 28.2% of all genes in the genome that have brain enriched expression. Gene ontology analyses also identified distinct differences in the biological process annotations of brain enriched genes within deleted and duplicated regions. Lastly we report for the first time 510 and 296 long non-coding RNAs found in deleted and duplicated regions respectively which have enriched expression in brain. Interestingly nine of our significant regions have only significant brain enriched lncRNA suggesting that for these regions the lncRNA is the most likely candidate for the observed phenotype. Our analyses highlight the diversity of genetic lesions that contribute to ASD and provide new genetic evidence for sub classification within the spectrum of ASD.