Loss of immune tolerance to autoantigens is a common feature of autoimmune diseases. Though these diseases share similar pathophysiology, and both genetic and environmental risk factors, no common mechanism for loss of tolerance has been robustly established. We begin this thesis by suggesting a new mechanism for loss of tolerance: via autoantigen loss of heterozygosity in the thymus. We then test a series of predictions from this theory. For proof of concept, we show that genetic variants which are overdominant (heterozygotes have higher disease risk than homozygotes) can be discovered in “major” autoantigen genes of four autoimmune diseases. Then, using a genome-wide association study (GWAS) of multiple sclerosis and set of its known autoantigen genes, we discover interactions between overdominant autoantigen variants. These interactions are consistent with a model where loss of tolerance to only one heterozygous autoantigen is necessary for disease. Loss of heterozygosity is the commonest form of somatic mutation and its rate increases with distance from a centromere: we find that mean overdominance among the multiple sclerosis autoantigen genes strongly correlates with this distance. We successfully replicate the interaction result using a held-out ulcerative colitis GWAS, and find a similar relationship between mean overdominance and genomic position. Finally, we model autoantigen loss of heterozygosity in the thymus and find it consistent with age of onset distributions for six diseases.