The quantitative real-time polymerase chain reaction (qRT-PCR) approach has become a widely used method to analyze expression patterns of target genes. The selection of an optimal reference gene is a prerequisite for the accurate normalization of gene expression in qRT-PCR. The present study constitutes the first systematic evaluation of potential reference genes in G. Don. Eleven candidate reference genes in different tissues and flowers at different developmental stages of were assessed using the following three software packages: GeNorm, NormFinder, and BestKeeper. The results showed that α (elongation factor 1-alpha), (18s ribosomal RNA), and (ribosomal protein L3) were the most stable reference genes in developing rhododendron flowers and, thus, in all of the tested samples, while tublin () was the least stable. (actin), , , and α were found to be the top four choices for different tissues, whereas was not found to favor qRT-PCR normalization in these tissues. Three stable reference genes are recommended for the normalization of qRT-PCR data in . Furthermore, the expression profiles of (phytoene synthase) and (phytoene dehydrogenase) were assessed using α, , , , and their combination as internals. Similar trends were found, but these trends varied when the least stable reference gene was used. The results further prove that it is necessary to validate the stability of reference genes prior to their use for normalization under different experimental conditions. This study provides useful information for reliable qRT-PCR data normalization in gene studies of .