Moser et al. report a new statistical method that is based on a hierarchical Bayesian mixture model (BayesR) for analysis of genome-wide association studies for human complex traits. Their method enables simultaneous discovery of variants associated with a complex trait, estimation of the total single-nucleotide polymorphism (SNP)-based genetic variance and polygenic risk prediction. BayesR is also used to characterize genetic architecture and partition the genetic variance across chromosomes. The authors tested their method in simulations and on case–control data sets for seven common diseases, finding that although a small proportion (<4%) of all SNPs contribute to a trait, a majority of these associated SNPs show a small effect, consistent with a polygenic model. The estimated genetic architecture varied considerably between traits, with type 1 diabetes and rheumatoid arthritis showing a greater proportion of the variance attributable to SNPs with larger effects, many of which are found in the major histocompatibility complex.