The focus of this paper is to illustrate and compare three recently suggested methods for the identification of differentially expressed genes and estimation of the false discovery rate (FDR) and to inversigate whether these FDR estimation methods are biased when we have a small sample size of 3 subjects per group. The methods are estimation of FDR based on averaging of fdr1d (FDR.avg), FDR estimation based on averaging of fdr2d (FDR.avg.fdr2d) and direct estimation of FDR using p-values (FDR.p). The properties are studied in a simulation experiment and a real data example. The comparison results shows that FDR.avg.fdr2d performs better than FDR.p and FDR.avg. The results further show the limitation of performance of the methods in very small sample size such as three subjects per group indicating the importance of developing a new FDR estimation procedure which works well under a situation of small sample size and unequal group variances.