Identification of differentially expressed (DE) genes with two or more conditions is an important task for discovery of few biomarker genes. Significance Analysis of Microarrays (SAM) is a popular statistical approach for identification of DE genes for both small- and large-sample cases. However, it is sensitive to outlying gene expressions and produces low power in presence of outliers. Therefore, in this paper, an attempt is made to robustify the SAM approach using the minimum β-divergence estimators instead of the maximum likelihood estimators of the parameters. We demonstrated the performance of the proposed method in a comparison of some other popular statistical methods such as ANOVA, SAM, LIMMA, KW, EBarrays, GaGa, and BRIDGE using both simulated and real gene expression datasets. We observe that all methods show good and almost equal performance in absence of outliers for the large-sample cases, while in the small-sample cases only three methods (SAM, LIMMA, and proposed) show almost equal and better performance than others with two or more conditions. However, in the presence of outliers, on an average, only the proposed method performs better than others for both small- and large-sample cases with each condition.