Polygenic risk scores (PRS) have potential utility as biomarkers of psychiatric disorders. However, while the schizophrenia (SZ) PRS has been consistently associated with case-control status and a more severe course of illness, the associations between the bipolar (BP) PRS and markers of bipolar disorder vary considerably between studies, with studies of population and case-control samples identifying many effects that cannot be replicated in case-only analyses. These analyses demonstrate that the heterogeneity in studies of the BP PRS is driven by selection bias. Specifically, selecting samples on the basis of diagnostic status or other phenotypes associated with genetic risk attenuates the correlation between the BP and SZ PRS. In such high-severity samples, while the SZ PRS predicts poor outcomes, the BP PRS predicts better outcomes. These findings highlight the importance of understanding the impact of selection bias in translational research evaluating PRS as biomarkers of psychiatric disorders, particularly when the intended application is populations enriched for high levels of genetic risk.