Using quantitative reverse transcription-polymerase chain reaction (RT-PCR), reference genes are utilized as endogenous controls for relative quantification of target genes in gene profiling studies. The suitability of housekeeping genes for that purpose in prostate cancer tissue has not been sufficiently investigated so far. The objective of this study was to select from a panel of 16 potential candidate reference genes the most stable genes for gene normalization. Expression of mRNA encoding ACTB, ALAS1, ALB, B2M, G6PD, GAPD, HMBS, HPRT1, K-ALPHA-1, POLR2A, PPIA, RPL13A, SDHA, TBP, UBC, and YWHAZ was examined in matched, microdissected malignant and nonmalignant tissue specimens obtained from 17 nontreated prostate carcinomas after radical prostatectomy by real-time RT-PCR. The genes studied displayed a wide expression range with cycle threshold values between 16 and 37. The expression was not different between samples from pT2 and pT3 tumors or between samples with Gleason scores <7 and >or=7 (P>0.05). ACTB, RPL13A, and HMBS showed significant differences (P<0.02 at least) in expressions between malignant and nonmalignant pairs. All other genes did not differ between the matched pairs, and the software programs geNorm and NormFinder were used to ascertain the most suitable reference genes from these candidates. HPRT1, ALAS1, and K-ALPHA-1 were calculated by both programs to be the most stable genes covering a broad range of expression. The expression of the target gene RECK normalized with HRPT1 alone and with the normalization factors generated by the combination of these three reference genes as well as with the unstable genes ACTB or RPL13A is given. That example shows the significance of using suitable reference genes to avoid erroneous normalizations in gene profiling studies for prostate cancer. The use of HPRT1 alone as a reference gene shown in our study was sufficient, but the normalization factors generated from two (HRPT1, ALAS1) or all three genes (HRPT1, ALAS1, K-ALPHA-1) should be considered for an improved reliability of normalization in gene profiling studies of prostate cancer.