Background: Whether the plasma proteome can predict the course of heart failure (HF) and has incremental value to established predictors is uncertain. Methods: Patients meeting Framingham HF criteria with history of reduced ejection fraction (n=1017) were prospectively enrolled in a registry and donated fasting blood samples. Plasma underwent analysis on the SOMAscan proteomic discovery platform, quantifying 4789 proteins using standard assay and quality controls. Patients were randomly divided into derivation (n=681) and validation (n=336) cohorts. We derived a proteomic risk score (PRS) in the derivation cohort using Lasso-penalized Cox regression and then tested it in the validation cohort. Both models were adjusted for an establish HF clinical risk score (MAGGIC) and NTproBNP. We assessed risk stratification improvement in the validation cohort by comparing models with and without PRS using the model C statistic, continuous net reclassification index (NRI), integrated discrimination index (IDI), and the median improvement in risk score (MIRS). Results: Overall 47.5% of patients were African American, 35.2% were female, mean ejection fraction was 34.8%, and average age was 67.9 years. After median follow-up of 3.6 years, there were 296 deaths (194 in derivation and 102 in validation). Optimized modeling defined a 32 protein PRS (hazard ratio [HR] 2.33, p<2.00E-16) which was also statistically significant when tested in the validation cohort (PRS HR=1.19, p=4.87E-02) and showed some improvement in risk stratification (Table). Conclusion: A plasma multi-protein predictive score can improve risk stratification in HF patients on top of a validated clinical score and NTproBNP. Additional investigation is warranted to define mechanisms underlying individual proteins and explore proteomic clinical applications.