Many common diseases show wide clinical or phenotypic variation. We present a statistical method for determining whether phenotypically defined subgroups of disease cases represent different genetic pathophysiologies, in which disease-associated variants have different effect sizes. Our method models the genomewide distributions of genetic association statistics with mixture Gaussians. We test for differential genetic bases without requiring explicit identification of disease-associated variants, maximizing power compared with standard variant-by-variant analyses. Where evidence for genetic subgroups is found, we present methods for subsequent identification of the contributing genetic variants. We demonstrate the method on simulated and test datasets where expected results are already known. We investigate subgroups of type 1 diabetes (T1D) cases defined by autoantibody positivity, establishing evidence for differential genetic basis with thyroid peroxidase antibody positivity. Our method determines the existence of a genetic basis for disease heterogeneity, enabling genomics to inform the development of precision medicine.