The number of ribs (NRs) and the carcass length (CL) are important economic traits. The traits are usually measured after slaughter. To improve the prediction performance of genomic selection (GS) for NRs and CL, one strategy is to integrate the significant loci identified from whole-genome sequencing (WGS) data by genome-wide association study (GWAS) into the genomic prediction (GP) model. This study investigated the GP of different genomic best linear unbiased prediction (GBLUP) and Bayesian models using chip genotype data, imputed WGS (iWGS) data and modeling significant single-nucleotide polymorphisms (SNPs) in different ways for the GP of NRs and CL in the Suhuai pig population. The prediction accuracy, bias and running time of 15 different GP models were evaluated by 10-fold cross-validation. The prediction accuracy of GBLUP using chip data for NRs and CL was 0.314 ± 0.022 and 0.194 ± 0.040, respectively. For NRs, based on the iWGS data, treating the most significant SNP as fixed effects in the GBLUP model had the highest predictive performance, with a prediction accuracy of 0.528 ± 0.023. For CL, based on the chip data, the model that added all the significant SNPs identified by imputed data by GWAS into the multi-trait GBLUP as the second random additive effect was the highest predictive performance, with a prediction accuracy of 0.305 ± 0.027. This study provides insights into optimizing GP models for small populations with phenotypes that are difficult to measure.