This article introduces PRS-CS, a novel polygenic prediction method that improves upon existing techniques by using Bayesian regression with continuous shrinkage priors on SNP effect sizes. PRS-CS leverages genome-wide association summary statistics and an external linkage disequilibrium (LD) reference panel to infer posterior effect sizes of SNPs. It employs a high-dimensional Bayesian framework and uses continuous shrinkage priors, which are robust to varying genetic architectures, offer computational advantages, and enable multivariate modeling of local LD patterns. Simulation studies using UK Biobank data show that PRS-CS outperforms existing methods across various genetic architectures, especially with large training samples. When applied to six common complex diseases and six quantitative traits in the Partners HealthCare Biobank, PRS-CS demonstrates improved prediction accuracy compared to alternative methods. PRS-CS is designed to work with GWAS summary statistics without requiring individual-level data, making it applicable in broader settings. The method uses a Bayesian regression framework with continuous shrinkage priors, allowing for adaptive shrinkage of SNP effect sizes based on their association strength. This approach enables efficient posterior inference and accurate modeling of local LD patterns. PRS-CS also includes a fully Bayesian version, PRS-CS-auto, which automatically learns the global shrinkage parameter from data. The method is computationally efficient and can be applied to large-scale data. Simulation studies and real-world applications show that PRS-CS consistently outperforms existing methods in prediction accuracy, particularly for highly polygenic traits. The method's ability to handle diverse genetic architectures and its computational efficiency make it a promising tool for polygenic prediction in clinical and research settings.This article introduces PRS-CS, a novel polygenic prediction method that improves upon existing techniques by using Bayesian regression with continuous shrinkage priors on SNP effect sizes. PRS-CS leverages genome-wide association summary statistics and an external linkage disequilibrium (LD) reference panel to infer posterior effect sizes of SNPs. It employs a high-dimensional Bayesian framework and uses continuous shrinkage priors, which are robust to varying genetic architectures, offer computational advantages, and enable multivariate modeling of local LD patterns. Simulation studies using UK Biobank data show that PRS-CS outperforms existing methods across various genetic architectures, especially with large training samples. When applied to six common complex diseases and six quantitative traits in the Partners HealthCare Biobank, PRS-CS demonstrates improved prediction accuracy compared to alternative methods. PRS-CS is designed to work with GWAS summary statistics without requiring individual-level data, making it applicable in broader settings. The method uses a Bayesian regression framework with continuous shrinkage priors, allowing for adaptive shrinkage of SNP effect sizes based on their association strength. This approach enables efficient posterior inference and accurate modeling of local LD patterns. PRS-CS also includes a fully Bayesian version, PRS-CS-auto, which automatically learns the global shrinkage parameter from data. The method is computationally efficient and can be applied to large-scale data. Simulation studies and real-world applications show that PRS-CS consistently outperforms existing methods in prediction accuracy, particularly for highly polygenic traits. The method's ability to handle diverse genetic architectures and its computational efficiency make it a promising tool for polygenic prediction in clinical and research settings.