Polygenic risk scores (PRS) have shown promise in predicting human complex traits and diseases. This paper introduces PRS-CS, a polygenic prediction method that uses genome-wide association summary statistics and an external linkage disequilibrium (LD) reference panel to infer posterior effect sizes of single nucleotide polymorphisms (SNPs). PRS-CS employs a high-dimensional Bayesian regression framework with continuous shrinkage (CS) priors on SNP effect sizes, which are robust to varying genetic architectures, provide substantial computational advantages, and enable multivariate modeling of local LD patterns. Simulation studies using UK Biobank data demonstrate that PRS-CS outperforms existing methods across a wide range of genetic architectures, especially with large training sample sizes. The method is applied to predict six common complex diseases and six quantitative traits in the Partners HealthCare Biobank, showing improved prediction accuracy over alternative methods. The continuous shrinkage priors allow for marker-specific adaptive shrinkage and accurate modeling of local LD patterns, enhancing the method's performance and computational efficiency.Polygenic risk scores (PRS) have shown promise in predicting human complex traits and diseases. This paper introduces PRS-CS, a polygenic prediction method that uses genome-wide association summary statistics and an external linkage disequilibrium (LD) reference panel to infer posterior effect sizes of single nucleotide polymorphisms (SNPs). PRS-CS employs a high-dimensional Bayesian regression framework with continuous shrinkage (CS) priors on SNP effect sizes, which are robust to varying genetic architectures, provide substantial computational advantages, and enable multivariate modeling of local LD patterns. Simulation studies using UK Biobank data demonstrate that PRS-CS outperforms existing methods across a wide range of genetic architectures, especially with large training sample sizes. The method is applied to predict six common complex diseases and six quantitative traits in the Partners HealthCare Biobank, showing improved prediction accuracy over alternative methods. The continuous shrinkage priors allow for marker-specific adaptive shrinkage and accurate modeling of local LD patterns, enhancing the method's performance and computational efficiency.