Quantitative real-time polymerase chain reaction (qRT-PCR) has been employed to study the gene expression profiles in human spermatozoa, but accurate analysis is dependent upon normalisation of data against an endogenous control. β-Actin (ACTB) and glyceraldehyde-3-phosphate dehydrogenase (GAPDH) are the most commonly used reference genes for normalisation of gene expression in human spermatozoa, but the expression of these genes in many tissues has considerable variation under different physiological, pathological and experimental conditions which limits their effectiveness in normalisation. The expression stability of a panel of 12 reference genes was studied in normal and pathological human spermatozoa using geNorm and NormFinder software. Although there were some discrepancies in the ranking of reference gene stability, each software program ranked B2M, ACTB, CYC1 and 18S RNA within the top 5 and recommended the combined use of at least two reference genes. Normalisation of qRT-PCR data for the cannabinoid receptor type 2 in spermatozoa using the different housekeeping genes demonstrated how, without validation, conflicting results are obtained. We recommend that the arbitrary use of reference genes should be avoided and the validation of reference gene stability should be undertaken prior to every study. For normalisation of CB2 expression, we would recommend using the geometric mean of B2M and ACTB.