Autism spectrum disorder (ASD) is a behaviorally defined condition that manifests in infancy or early childhood as deficits in communication skills and social interactions. Often, restricted and repetitive behaviors (RRBs) accompany this disorder. ASD is polygenic and genetically complex, so we hypothesized that focusing analyses on intermediate core component phenotypes, such as RRBs, can reduce genetic heterogeneity and improve statistical power. Applying this approach, we mined Caucasian genome-wide association studies (GWAS) data from two of the largest ASD family cohorts, the Autism Genetics Resource Exchange and Autism Genome Project (AGP). Of the 12 RRBs measured by the Autism Diagnostic Interview-Revised, seven were found to be significantly familial and substantially variable, and hence, were tested for genome-wide association in 3104 ASD-affected children from 2045 families. Using a stringent significance threshold (P<7.1 × 10-9), GWAS in the AGP revealed an association between 'the degree of the repetitive use of objects or interest in parts of objects' and rs2898883 (P<6.8 × 10-9), which resides within the sixth intron of PHB. To identify the candidate target genes of the associated single-nucleotide polymorphisms at that locus, we applied chromosome conformation studies in developing human brains and implicated three additional genes: SLC35B1, CALCOCO2 and DLX3. Gene expression, brain imaging and fetal brain expression quantitative trait locus studies prioritize SLC35B1 and PHB. These analyses indicate that GWAS of single heritable features of genetically complex disorders followed by chromosome conformation studies in relevant tissues can be successful in revealing novel risk genes for single core features of ASD.