Numerous algorithms exist for producing gene sets from high-throughput genomic and proteomic technologies. However, analysis of the functional significance of these groups of genes or proteins remains a big challenge. We developed a Web based system called gene-set cohesion analysis tool (GCAT) for estimating the significance level of the functional cohesion of a given gene set. The method utilizes latent semantic indexing (LSI) derived gene-gene literature similarities to determine if the functional coherence of a gene set is statistically significant compared to that expected by chance. The robustness of the method was determined by evaluating the functional cohesion for over 6000 gene ontology categories. Here, we demonstrate the utility of GCAT for analysis of microarray data from previously published experiments in which embryonic fibroblasts were treated with interferon. Using GCAT, we found the highest literature cohesion p-value (p= 1.37E-63) corresponded to a set of genes that were differentially regulated > 2-fold and had a t-test p-value <0.05, compared to genes that were only changed >2-fold (literature p-value=2.2E-44) or had a p-value <0.05 (literature p-value=6.0E-32). As a control, genes that were changed less than 2-fold or had a p-value >0.05 did not show a significant literature cohesion. These results demonstrate that GCAT can provide an objective literature-based measure to evaluate the biological significance of gene sets identified by different criterions. GCAT is available at http://motif.memphis.edu/gcat/.