Traditionally, autistic spectrum disorder (ASD) and specific language impairment (SLI) are regarded as distinct conditions with separate etiologies. Yet these disorders co-occur at above chance levels, suggesting shared etiology. Simulations, however, show that additive pleiotropic genes cannot account for observed rates of language impairment in relatives, which are higher for probands with SLI than for those with ASD + language impairment. An alternative account is in terms of 'phenomimicry', i.e., language impairment in comorbid cases may be a consequence of ASD risk factors, and different from that seen in SLI. However, this cannot explain why molecular genetic studies have found a common risk genotype for ASD and SLI. This paper explores whether nonadditive genetic influences could account for both family and molecular findings. A modified simulation involving G x G interactions obtained levels of comorbidity and rates of impairment in relatives more consistent with observed values. The simulations further suggest that the shape of distributions of phenotypic trait scores for different genotypes may provide evidence of whether a gene is involved in epistasis.