Genome-wide evaluation uses the associations of a large number of single nucleotide polymorphism (SNP) markers across the whole genome and then combines the statistical methods with genomic data to predict the genetic values. Genomic predictions relieson linkage disequilibrium (LD) between genetic markers and quantitative trait loci (QTL) in a population. Methods that use all markers simultaneously may therefore result in greater reliabilities of predictions of the total genetic merit, indicating that a larger proportion of the genetic variance is explained. This is hypothesized that the genome-wide methods deal differently with genetic architecture of quantitative traits and genome. The genomic nonlinear Bayesian variable selection methods (BayesA, BayesBand Bayesian LASSO) are compared using the stochastic simulation across three effective population sizes (Ne). Thereby, a genome with three chromosomes, 100 cM each was simulated. For each animal, a trait was simulated with heritability of 0.50, three different marker densities (1000, 2000 and 3000 markers) and number of the QTL was assumed to be either 100, 200 or 300. The data were simu-lated with two different QTL distributions which were uniform and gamma (α=1.66, β=0.4). Marker den-sity, number of the QTL and the QTL effect distributions affected the genomic estimated breeding value accuracy with different Ne (P<0.05). In comparison of three methods, the greatest genomic accuracy ob-tained by BayesB method for traits influenced by a low number of the QTL, high marker density, gamma QTL distribution and high Ne.