Several reference genes have been used to quantify gene expression in human epilepsy surgery tissue. However, their reliability has not been validated in detail, although this is crucial in interpreting epilepsy-related changes of gene expression. We evaluated 12 potential reference genes in neocortical tissues resected from patients with temporal lobe epilepsy (TLE) with either few or many seizures (n=6 each) and post mortem controls (n=6) using geNorm and NormFinder algorithms. For all candidate reference genes threshold cycle (C(T)) values were measured. geNorm analysis revealed that the expression of e.g. glyceraldehyde-3-phosphate-dehydrogenase (GAPDH) and hypoxanthine phosphoribosyl-transferase (HPRT) is unstable, whereas synaptophysin (SYP) and neuron-specific enolase (NSE)/mitochondrial 39S ribosomal protein L28 (MRPL) are most stably expressed. The geometric mean of SYP, NSE and MRPL levels is recommended as normalization factor (NF). NormFinder analysis, in contrast, indicated HPRT as the most stable single gene and recommended the geometric mean of TATA-box binding protein (TBP) and NSE levels as NF. Different values of upregulation of glial fibrillary protein (GFAP) expression were found in TLE tissue compared to control tissue depending on the NF used: 4.5-fold (geNorm-NF), 4.7-fold (NormFinder-NF), 4.2-fold (vs. GAPDH) and 7.8-fold (vs. HPRT). The expression of GABA(A) receptor subunit α5 (GARα5) was unaltered in the TLE groups compared to controls (geNorm-NF, NormFinder-NF, vs. GAPDH). However, normalization to HPRT suggests an apparent increase of GARα5 expression. In conclusion, the geNorm-NF (SYP/NSE/MRPL) and the NormFinder-NF (TBP/NSE) are equally suitable for normalization of gene expression in the human epileptogenic neocortex. In contrast, normalization to single and probably less stably expressed genes may not deliver accurate results.