The pathology of Progressive Supranuclear Palsy (PSP) causes Richardson’s syndrome (RS) and variant clinical phenotypes, with differential cognitive, behavioural and motor deficits. Survival is 3-4 years from diagnosis. The PSP Rating Scale (PSPRS) is prognostically informative, but the impact of cognitive and behavioural changes on survival is less clear. We test univariate and multivariate models of survival to determine the best clinically-applicable model for all-cause mortality.The MDS 2017 criteria were used to phenotype patients at the Cambridge Centre for Parkinson-plus (UK). Univariate and multivariate logistic regression models assessed the relationship between survival and clinicalvariables(PSPRS, MMSE, Addenbrooke’s Cognitive Examination,Cambridge Behavioural Inventory).335 people (male=56%, age 71.4±7.2 years) were identified with possible, probable or definite PSP. RS and variant groups had similar disease severity at baseline assessment (p=0.6) and survival (p=0.2). For 3-year mortality, PSPRS was the most reliable single predictor (AUC=0.68). Age, sex and PSPRS improved the model(AUC=0.71), but over all models Akaike’s Information Criterion identified the best model for RS to include PSPRS, CBI and MMSE(AUC=0.79, p=0.01). CBI and MMSE also improved the model for var- iant-PSP(PSPRS, CBI and MMSE AUC=0.89 vs 0.73, p=0.01).Inclusion of cognitive and behavioural measures improves the prediction of mortality in PSP.