Several strategies including GBLUP and ssGBLUP methods have been proposed to predict genomic breeding values (GEBV) for dairy cattle. Many studies have reported that the accuracy of GEBV predicted from ssGBLUP was higher than from GBLUP. There is no report applying same analysis to Japanese Holsteins. The objective of this study was to compare the accuracies of GPI predicted by GBLUP and ssGBLUP. The traits used in this analysis were 305-d milk, fat and protein yields, and feet and legs, udder and final score. In total, 3,787 bulls genotyped by the Illumina BovineSNP50 BeadChip were used. Validation bulls were prediction bulls with no daughters in 2009 data but more than 20 daughters in 2013 data. In multistep evaluation, training bulls were required to have 10 daughters in 2009 data. Number of training bulls were 517 and 554 for the analysis of milk and type traits, respectively. For multistep evaluation, GPI were estimated by selection index blending after calculating DGV by linear GBLUP. Coefficients of determination (R²) by regression analysis of GPI for 2009 data on deregressed EBV for 2013 data were used as indicator of accuracy. The R² ranged from 0.19 (protein) to 0.27 (milk) and from 0.17 (feet and legs) to 0.28 (milk) for DGV and GPI predicted by GBLUP and from 0.21 (protein) to 0.30 (milk) for GEBV predicted by ssGBLUP, respectively. Hence, it is suggested that the GEBV predicted by ssGBLUP is higher accuracy than DGV and GPI predicted by GBLUP.