The selection of most appropriate reference gene(s) is a crucial step in studies quantifying gene expression by quantitative real-time PCR (qPCR). As a need reference gene(s) should be unaffected at transcription level by experimental conditions and/or tissue types. In this study, 11 candidate reference genes were tested in Clarias batrachus for their expression stability using the geNorm, NormFinder and BestKeeper statistical algorithms, and were compared in different tissues (brain, heart, liver, muscle, spleen and head kidney) and treatments (before and after hypoxia exposure). The results indicated that in brain, heart, liver, muscle, spleen and head kidney, respectively, 28S rRNA/TUB, 28S rRNA/TUB, RPL30/ 28S rRNA, RPL30/TUB, ELF-1A/28S rRNA and ELF-1A/TUB gene pairs were highly stable and were suitable as reference genes to study oxidative stress, while ACTB and B2M were the least stable genes in examined tissues under normoxic and/or hypoxic conditions. The observations suggested the consideration of tissue types and use of at least two reference genes, instead of one in accurate normalization of qPCR data.