Progression of prostate cancer during endocrine therapy is a major clinical problem, the molecular mechanisms of which remain poorly understood. Amplification of the androgen receptor (AR) gene was recently described in recurrent prostate carcinomas from patients who had failed androgen deprivation therapy. To evaluate the hypothesis that amplification of the AR gene is a cause for the failure of androgen deprivation therapy in prostate cancer, we studied whether AR amplification leads to gene overexpression, whether the amplified AR gene is structurally intact, and whether tumors with AR amplification have distinct biological and clinical characteristics. Tumor specimens were collected from 54 prostate cancer patients at the time of a local recurrence following therapy failure. In 26 cases, paired primary tumor specimens from the same patients prior to therapy were also available. Fifteen (28%) of the recurrent therapy-resistant tumors, but none of the untreated primary tumors, contained AR gene amplification as determined by fluorescence in situ hybridization. According to single-stranded conformation polymorphism analysis, the AR gene was wild type in all but one of the 13 AR amplified cases studied. In one tumor, a presumed mutation in the hormone-binding domain at codon 674 leading to a Gly --> Ala substitution was found, but functional studies indicated that this mutation did not change the transactivational properties of the receptor. AR amplification was associated with a substantially increased level of mRNA expression of the gene by in situ hybridization. Clinicopathological correlations indicated that AR amplification was most likely to occur in tumors that had initially responded well to endocrine therapy and whose response duration was more than 12 months. Tumors that recurred earlier or those that showed no initial therapy response did not contain AR amplification. The median survival time after recurrence was two times longer for patients with AR amplification in comparison to those with no amplification (P = 0.03, Willcoxon-Breslow test). In conclusion, failure of conventional androgen deprivation therapy in prostate cancer may be caused by a clonal expansion of tumor cells that are able to continue androgen-dependent growth despite of the low concentrations of serum androgens. Amplification and the increased expression of a wild-type AR gene may play a key role in this process.