We analyzed the polymorphic (CAG)n and (GGN)n regions within the androgen receptor gene from participants in a population-based case-control study of prostate cancer in middle-aged (40-64 years) Caucasian men. The associations between repeat lengths and risk of prostate cancer and the effects of confounding and modifying factors, such as age, family history of prostate cancer, and body mass index, were evaluated. DNA was available for 301 cases and 277 controls. The overall age-adjusted relative odds of prostate cancer associated with the number of (CAG) repeats as a continuous variable was 0.97 [95% confidence interval (CI), 0.92-1.03], suggesting a 3% decrease in risk of prostate cancer for each additional (CAG) repeat. Further analyses identified several subgroups at increased risk. These were men with less than the median number of CAG repeats (< 22) that were younger [< 60 years; relative odds (RO), 1.47; 95% CI, 0.96-2.25], had an affected first-degree relative (RO, 1.59; 95% CI, 0.62-4.14), or were relatively thin (Quetelet index < 24.4; RO, 2.21; 95% CI, 1.07-4.69). Although only the latter result was statistically significant, these results are provocative and support the hypothesis that (CAG)n array length is a predictor of risk for prostate cancer. Similar analyses of (GGN)n showed that with the exception of men with a family history of prostate cancer and those in the highest quartile of body mass index, men with < or = 16 repeats had higher risk estimates than did men with > 16 repeats. Overall, those men who had < or = 16 repeats had a significant elevation in risk (RO, 1.60; 95% CI, 1.07-2.41). When both repeat lengths were considered jointly, the subgroup with two short repeats (CAG, < 22; GGN, < or = 16) had a 2-fold elevation in odds (RO, 2.05; 95% CI, 1.09-3.84) relative to those with two long repeats (CAG, > or = 22; GGN, > 16). These data suggest that determination of both androgen receptor repeats within germ-line DNA may be useful in assessing an individual's risk of developing prostate cancer.