Opposite phenotypic and behavioural traits associated with copy number variation and disruptions to imprinted genes with parent-of-origin effects have led to the hypothesis that autism and schizophrenia share molecular risk factors and pathogenic mechanisms, but a direct phenotypic comparison of how their risks covary has not been attempted. Here, we use health registry data collected on Denmark's roughly 5 million residents between 1978 and 2009 to detect opposing risks of autism and schizophrenia depending on normal variation (mean ± 1 s.d.) in adjusted birth size, which we use as a proxy for diametric gene-dosage variation in utero . Above-average-sized babies (weight, 3691–4090 g; length, 52.8–54.3 cm) had significantly higher risk for autism spectrum (AS) and significantly lower risk for schizophrenia spectrum (SS) disorders. By contrast, below-average-sized babies (2891–3290 g; 49.7–51.2 cm) had significantly lower risk for AS and significantly higher risk for SS disorders. This is the first study directly comparing autism and schizophrenia risks in the same population, and provides the first large-scale empirical support for the hypothesis that diametric gene-dosage effects contribute to these disorders. Only the kinship theory of genomic imprinting predicts the opposing risk patterns that we discovered, suggesting that molecular research on mental disease risk would benefit from considering evolutionary theory.