The relative concentration of fatty acids within soybean oil is of major interest to consumers of soybean products.  Marker-assisted breeding strategies are being pursued to improve the fatty acid composition of soybean oil in a timely and resource efficient manner.  One such strategy that may be useful for fatty acid improvement is genomic selection, as fatty acids exhibit quantitative inheritance.  The purpose of this research was to create a genomic selection training set to test the effectiveness of this method for fatty acid improvement in a soybean population consisting of 887 F5 derived recombinant inbred lines (RIL).  This population was genotyped with over 17,000 polymorphic SNP markers spread throughout the soybean genome.  In order to simulate progeny rows, each RIL was grown in a single plot in 2010 in Knoxville, Tennessee.  The fatty acid composition of each plot was measured using gas chromatography (GC). The combined genotypic and phenotypic data for 588 RIL from the 2010 field test was used to generate a genomic selection training set for each soybean fatty acid.  Using this training set, fatty acid phenotypes from the remaining 299 RIL were estimated from their respective genotypes.  Correlations of the rankings between genotype estimated and GC measured phenotype for the 299 RIL ranged from 0.41 to 0.68 (P < 0.0001) for the five fatty acids.  In order to compare the effectiveness of making genomic selections within the 299 RIL, comparisons were made between estimated and measured phenotypes for 10%, 15%, and 20% selection intensity.  Matches between selection methods ranged from 23.3-43.3% (10% selection), 31.1-55.6% (15% selection), and 40-58.3% (20% selection).  As evidenced by this data, relaxation of selection intensity leads to greater selection agreement between methods.  Future studies using this population will provide further insight into the value of genomic selection for soybean fatty acids.