Abstract Importance Bipolar disorder (BD) overlaps schizophrenia in its clinical presentation and genetic liability. Alternative approaches to patient stratification beyond current diagnostic categories are needed to understand the underlying disease processes/mechanisms. Objectives To investigate the relationship between common-variant liability for schizophrenia, indexed by polygenic risk scores (PRS) and psychotic presentations of BD, using clinical descriptions which consider both occurrence and level of mood-incongruent psychotic features. Design Case-control design: using multinomial logistic regression, to estimate differential associations of PRS across categories of cases and controls. Settings & Participants 4399 BDcases, mean [sd] age-at-interview 46[12] years, of which 2966 were woman (67%) from the BD Research Network (BDRN) were included in the final analyses, with data for 4976 schizophrenia cases and 9012 controls from the Type-1 diabetes genetics consortium and Generation Scotland included for comparison. Exposure Standardised PRS, calculated using alleles with an association p-value threshold < 0.05 in the second Psychiatric Genomics Consortium genome-wide association study of schizophrenia, adjusted for the first 10 population principal components and genotyping-platform. Main outcome measure Multinomial logit models estimated PRS associations with BD stratified by (1) Research Diagnostic Criteria (RDC) BD subtypes (2) Lifetime occurrence of psychosis.(3) Lifetime mood-incongruent psychotic features and (4) ordinal logistic regression examined PRS associations across levels of mood-incongruence. Ratings were derived from the Schedule for Clinical Assessment in Neuropsychiatry interview (SCAN) and the Bipolar Affective Disorder Dimension Scale (BADDS). Results Across clinical phenotypes, there was an exposure-response gradient with the strongest PRS association for schizophrenia (RR=1.94, (95% C.I. 1.86, 2.01)), then schizoaffective BD (RR=1.37, (95% C.I. 1.22, 1.54)), BD I (RR= 1.30, (95% C.I. 1.24, 1.36)) and BD II (RR=1.04, (95% C.I. 0.97, 1.11)). Within BD cases, there was an effect gradient, indexed by the nature of psychosis, with prominent mood-incongruent psychotic features having the strongest association (RR=1.46, (95% C.I. 1.36, 1.57)), followed by mood-congruent psychosis (RR= 1.24, (95% C.I. 1.17, 1.33)) and lastly, BD cases with no history of psychosis (RR=1.09, (95% C.I. 1.04, 1.15)). Conclusion We show for the first time a polygenic-risk gradient, across schizophrenia and bipolar disorder, indexed by the occurrence and level of mood-incongruent psychotic symptoms.