Reverse transcription-polymerase chain reaction (RT-PCR) approaches have been used in a large proportion of transcriptome analyses published to date. The accuracy of the results obtained by this method strongly depends on accurate transcript normalization using stably expressed genes, known as references. Statistical algorithms have been developed recently to help validate reference genes, and most studies of gene expression in mammals, yeast and bacteria now include such validation. Surprisingly, this important approach is under-utilized in plant studies, where putative housekeeping genes tend to be used as references without any appropriate validation. Using quantitative RT-PCR, the expression stability of several genes commonly used as references was tested in various tissues of Arabidopsis thaliana and hybrid aspen (Populus tremula × Populus tremuloides). It was found that the expression of most of these genes was unstable, indicating that their use as references is inappropriate. The major impact of the use of such inappropriate references on the results obtained by RT-PCR is demonstrated in this study. Using aspen as a model, evidence is presented indicating that no gene can act as a universal reference, implying the need for a systematic validation of reference genes. For the first time, the extent to which the lack of a systematic validation of reference genes is a stumbling block to the reliability of results obtained by RT-PCR in plants is clearly shown.