Insoybean [Glycine max (L.) Merr.], seed oil concentration is a complex quantitative trait, and genomic selection (GS) has been shown to be a valuable tool for performing selection on such traits. The objectives of this study were to evaluate multiple GS models for seed oil concentration using a low-density marker panel in four biparental soybean populations and to assess predictive ability of the models using six unique training populations (TPs). Individuals were grown as BC 1 F 4 :F 5 progeny rows in 2014. Genomic estimated breeding values (GEBVs) were calculated for each genotype within a population using genomic best linear unbiased predictor (GBLUP), BayesA, and BayesB models in a biparental specific context. In 2015, 60 individuals from each population were randomly selected and grown at six locations with two replications each to generate a "true" phenotypic value for each genotype. Prediction accuracies for each estimation set were generated by correlating the GEBVs with the "true" phenotypic value. Across all populations, the GS prediction accuracy was greatest using GBLUP; however, no GS prediction model showed a significant advantage in accuracy over the phenotypic values. Generally, TPs consisting of more individuals had higher prediction accuracies; however, variations were observed across populations and models. The results show that GS in a biparental context and with low marker densities can be a valuable tool for breeders focused on making gains for oil concentration; however, consideration must be given as to how to apply these methods to each situation.