Considerations when using a reference gene Reference gene validation exercises are also subject to the problem of normalisation. The strategy we previously presented5 uses total RNA to normalise the sample prior to reference gene variability assessment. The different reference genes are then measured by real-time RT-PCR and variation in the cycle threshold (Ct) or crossing point (Cp) assessed. As RNA normalisation can incorporate error and does not take into account the RT step, the measured reference gene variability represents the cumulative error of the entire process, that is, the innate variation of the reference gene under investigation and the experimental error associated with the technique. Once this variation is defined the chosen reference gene can provide the resolution of the assay in question. Choosing the accepted level of variability will depend on the degree of resolution required. Even if the chosen gene is variable it may not matter as long as intergroup difference being measured is greater than the reference gene variation, that is, a reference gene RNA that has an error of 1 log may not be ideal, but is sufficient to measure a 2 log change in a gene of interest. There are a number of programs based on the excel platform that allow the assessment of multiple reference genes. Gnorm allows the most appropriate reference gene to be chosen by using the geometric mean of the expression of the candidate cDNA.26 This software is freely available (http://www.genomebiology.com/ 2002/3/7/research/0034/) and the underlying principles are published by Vandesompele et al.26 BestKeeper also selects the least variable gene using the geometric mean but uses raw data27 instead of data converted to copy number, it is also available at http://www.genequantification.de/BestKeeper-1.zip. A third program Norm-Finder,28 freely available on request, not only measures the variation but also ranks the potential reference genes by how much they differ between study groups, that is, the extent by which they are effected by the experimental conditions. Defining this is essential as it can generate false results as discussed above. Vandesompele et al also advocate the use of multiple reference genes rather than relying on a single RNA transcript. This is a robust method for providing accurate normalisation and is consequently favourable if fine measurements are to be made. However, it is not always possible to measure multiple reference genes due to limited sample availability and cost. Furthermore, even if multiple genes are chosen the resolution of the particular assay remains dependent on the variability of the chosen reference genes.