In quantitative reverse transcription-polymerase chain reaction (qRT-PCR), normalization using reference genes is a common useful approach, but the validation of suitable reference genes remains a crucial problem. Use of unconfirmed reference genes may lead to misinterpretation of the expression of target genes. The aim of this study was to adapt an adequate statistical approach to identify and validate reference genes suitable for normalization in qRT-PCR assays. We introduce the equivalence test for the identification of stably expressed reference genes. To evaluate the advantages of this test, the expression of five genes widely used as reference genes (18S, B2M, HPRT1, LMNB1, and SDHA), and of two target genes (TP53 and MMP2), was determined with qRT-PCR in different tissues (clear cell renal cell carcinoma, colon carcinoma, and gastrointestinal stromal tumors). We demonstrate that a stable expression of a reference gene in one tumor type does not predict a stable expression in another tumor type. In addition, we found that even within one tumor type, the expression of a reference gene was not stable for different biological groupwise comparisons. These observations confirm that there is no universal reference gene and underline the importance of specific validation of potential reference genes for any experimental condition.