Genome-wide association studies (GWAS) have identified numerous genetic loci associated with brain disorders, yet these explain just a small proportion of the heritability. Polygenicity hampers the detection of additional risk genes in practical sample sizes. In this talk, I will introduce a novel approach, which combines knowledge from animal models, evolutionary genomics as well as biochemical information on the genome, to find genes for human complex disorders. In the first part, I will introduce the dog as a natural model, and explain how its unique population history (old domestication + recent breed creation) has made modern purebred dogs as excellent natural models for human diseases. I will show how we found genes associated with canine model of human obsessive-compulsive disorder (OCD) using GWAS and targeted resequencing of affected and healthy dogs (Tang, Noh et al. Genome Biology 2014). In the second part, I will introduce a focused genetic search for human OCD risk loci. We compiled a list of 608 genes implicated by studies in dogs, a natural model for OCD, mice, an artificial model, and in humans. We resequenced the coding and conserved non-coding regions in 592 OCD patients and 560 controls. Using a new analysis method that incorporates evolutionary and biochemical annotations, we identify five significantly associated genes despite the modest cohort size. We genotyped the top candidate variants in an additional 1,834 individuals (combined: 1,298 OCD cases and 1,660 controls), and find strong gene associations implicating abnormal synapse adhesion and maintenance in OCD. In NRXN1 (p=7.3×10−7), we find 7 missense variants that alter protein-binding domains interacting with its post-synaptic partners. In CTTNBP2 and REEP3, two genes involved in synapse maintenance and cellular vesicle trafficking, we find more than 20 regulatory variants, at least one third of which alter transcription factor-DNA binding domains. For a polygenic trait like OCD, an animal model-driven approach, combined with an analysis method incorporating evolutionary and regulatory information, can identify associated genes and candidate variants, both coding and non-coding, in a much smaller sample cohort than conventional approaches.