Mood disorders, including major depression and bipolar disorder, represent a substantial public health burden. Genome-wide association studies (GWAS) of mood disorders in large case-control cohorts have identified numerous loci that together account for part of the risk, but pathophysiological mechanisms have been elusive. Individual common variants confer very small risk, and samples have been underpowered to detect rare variants with larger effects. Extended pedigrees from population isolates could address these challenges, as some risk-conferring alleles may be highly enriched, relative to the broader population. Here I report a GWAS of mood disorders in an Old Order Amish (OOA) founder population. By integrating genetic and phenotypic data from three independently collected OOA cohorts enriched for mood disorders with OOA population controls (total n=1,672), I identified four genome-wide significant risk loci, 2 of which are novel. The observed effect sizes of the associated loci are substantially greater than those identified in the broader population, and the risk-associated haplotypes harbor otherwise rare coding variants enriched in OOA. At one locus, I identified a population-enriched, non-synonymous variant in the CUX1 alternative splicing product associated with >3-fold relative risk. Quantitative behavioral and neurocognitive traits in a sub-set of 314 subjects revealed effects of risk variants on sub-clinical depressive symptoms and working memory. These findings provide insight into the genetic architecture of mood disorders and provide a substrate for mechanistic and clinical studies.