Real-time reverse transcriptase-polymerase chain reaction (RT-PCR) represents a sensitive and efficient technique to determine expression levels of target genes in multiple samples and is increasingly used in clinical oncology to evaluate the patient's outcome or to detect minimal residual disease. Normalization of raw data are required to obtain comparable results between different specimens and is usually achieved by correlating transcript abundances of target genes with those of a single control gene with putatively stable expression levels. In this study, expression stability of six supposed control genes was evaluated in 64 samples of primary neuroblastoma and HPRT1 and SDHA mRNA levels were shown to exhibit the least expression variability among the samples. Because application of more than one control gene may enhance reliability of real-time RT-PCR results, various normalization factors consisting of the geometrical mean of multiple control gene expression values were calculated and evaluated by mRNA quantification of 14 target genes. Comparison with transcript levels determined by oligonucleotide-array expression analysis revealed that target gene mRNA quantification became most consistent after normalization to averaged expression levels of HPRT1 and SDHA. This normalization factor was in addition demonstrated to be not associated with stage of disease or MYCN amplification status of the tumor. Thus, these data indicate that the geometrical mean of HPRT1 and SDHA transcript levels represents a suitable internal control for biological and clinical studies investigating differential gene expression in primary neuroblastoma by real-time RT-PCR.