Real-time polymerase chain reaction (PCR) is commonly used for a sensitive and specific quantification of messenger RNA (mRNA). The levels of mRNA are frequently compared between two or more experimental groups. However, such comparisons require normalization procedures, and reference genes are frequently used for this purpose. We discuss pitfalls in normalization and specifically in the choice of reference genes. Reference genes, which prove suitable for some experimental conditions, are not necessarily similarly appropriate for others. Therefore,a proper validation of the suitability of a given reference gene or sets thereof is required for each experimental setting. Several computer programmes are available to aid such validation.