Introduction: Estimating the genetic risk of developing complex diseases, such as coronary artery disease (CAD), is now possible by aggregating data from genome-wide association studies (GWAS) into polygenic risk scores (PRS). This study aims to evaluate the performance of PRS derived from CAD (PRS CAD ) and blood lipids GWAS to predict severe CAD defined as myocardial infarction or coronary revascularization (MI-REVASC). Methods: Using nuclear magnetic resonance data from UK Biobank (UKB, n=90300), we conducted 29 GWAS for blood lipids (9 for fatty acids and 20 for lipoproteins). Summary statistics of these GWAS were then used to derive PRS in the rest of UKB (n=318422, median follow-up = 12.6 years). A PRS CAD was also calculated using the publicly available GWAS from the CARDIoGRAMplusC4D consortium. All PRS included >1.1 million of genetic variants and were calculated using LDpred2. Hazard Ratio (HR) for incident MI-REVASC (n=10935) are reported with their 95% CIs, for 1-SD increase of each PRS. Selected independent PRS were combined into multivariate Cox regression models, for which predictive performance was evaluated. All models were adjusted for age, sex and genetic principal components. Results: Out of 30 PRS, 28 were significantly associated with incident MI-REVASC; PRS CAD showed the strongest association (HR=1.60 [1.57-1.63], p<10e-300), whereas PRS for small LDL particles (sLDL-P) was the most associated blood lipids PRS (HR=1.14 [1.12-1.16], p=7.63e-42). The optimal model included PRS for CAD, sLDL-P, HDL-C and the PUFA/MUFA (poly-/mono- unsaturated fatty acids) ratio. Discriminative capacities for incident MI-REVASC were significantly increased in the model including PRS CAD (C-statistic (C) = 0.744 [0.739-0.748]) compared to the model without PRS (C= 0.710 [0.705-0.714]). No significant difference was observed when independent blood lipids PRS were added to the model (C= 0.745 [0.741 to 0.749]). Conclusions: Although independently associated with the incidence of severe CAD, blood lipids PRS provide little improvement in the predictive performance when added to PRS CAD .