Androgen receptor (AR) has long been hypothesized to play an important role in prostate cancer etiology. Two trinucleotide repeat polymorphisms (CAG and GGC repeats in exon 1 of the AR gene) have been investigated as risk factors for prostate cancer in several studies. However, the results are inconclusive, probably because of the variations of study designs, characteristics of study samples, and choices of analytical methods. In this study, we evaluated evidence for linkage and association between the two AR repeats and prostate cancer by using the following comprehensive approaches: (1) a combination of linkage and association studies, (2) a test for linkage by parametric analysis and the male-limited X-linked transmission/disequilibrium test (XLRC-TDT), (3) a test for association by using both population-based and family-based tests, and (4) a study of both hereditary and sporadic cases. A positive but weak linkage score (HLOD=0.49, P=0.12) was identified in the AR region by parametric analysis; however, stronger evidence for linkage in the region, especially at the GGC locus, was observed in the subset of families whose proband had < or = 16 GGC repeats (HLOD=0.70, P=0.07) or by using XLRC-TDT ( z'=2.65, P=0.008). Significantly increased frequencies of the < or = 16 GGC repeat alleles in 159 independent hereditary cases (71%) and 245 sporadic cases (68%) cases compared with 211 controls (59%) suggested that GGC repeats were associated with prostate cancer ( P=0.02). Evidence for the association between the < or = 16 GGC repeats and prostate cancer risk was stronger with XLRC-TDT ( z'=2.66, P=0.007). No evidence for association between the CAG repeats and prostate cancer risk was observed. The consistent results from both linkage and association studies strongly implicate the GGC repeats in the AR as a prostate cancer susceptibility gene. Further studies on this polymorphism in other independent data sets and functional analysis of the GGC repeat length on AR activity are warranted.