Background: Real-time reverse transcriptase quantitative polymerase chain reaction (real-time RTqPCR) is a technique used to measure mRNA species copy number as a way to determine key genes involved in different biological processes. However, the expression level of these key genes may vary among tissues or cells not only as a consequence of differential expression but also due to different factors, including choice of reference genes to normalize the expression levels of the target genes; thus the selection of reference genes is critical for expression studies. For this purpose, ten candidate reference genes were investigated in bovine muscular tissue. Results: The value of stability of ten candidate reference genes included in three groups was estimated: the so called 'classical housekeeping' genes (18S, GAPDH and ACTB), a second set of genes used in expression studies conducted on other tissues (B2M, RPII, UBC and HMBS) and a third set of novel genes (SF3A1, EEF1A2 and CASC3). Three different statistical algorithms were used to rank the genes by their stability measures as produced by geNorm, NormFinder and Bestkeeper. The three methods tend to agree on the most stably expressed genes and the least in muscular tissue. EEF1A2 and HMBS followed by SF3A1, ACTB, and CASC3 can be considered as stable reference genes, and B2M, RPII, UBC and GAPDH would not be appropriate. Although the rRNA-18S stability measure seems to be within the range of acceptance, its use is not recommended because its synthesis regulation is not representative of mRNA levels. Conclusion: Based on geNorm algorithm, we propose the use of three genes SF3A1, EEF1A2 and HMBS as references for normalization of real-time RTqPCR in muscle expression studies.