A longstanding goal has been to determine whether androgen receptor (AR) levels could be used to predict the clinical response of metastatic prostate cancer to androgen withdrawal therapy. A major limitation of previous studies was the use of homogenized tissue, which yields an average AR content for all cells. By AR immunohistochemical study using an antibody specific for AR the authors assessed nuclear AR content specifically in the malignant epithelial cells of prostate needle biopsy specimens of 17 patients with Stage D prostate cancer. The authors found that prostate cancer contains AR‐positive and AR‐negative malignant cells before androgen withdrawal therapy, but the percentage of AR‐positive cells did not predict the time to tumor progression after therapy. There was no significant correlation between the percentage of AR‐positive malignant cells and the time to tumor progression. When patients were divided into two groups based on the median time to progression, the percentage of AR‐positive nuclei was not significantly different in poor responders versus good responders. When patients were divided into two groups based on the median percentage of receptor‐positive nuclei, Kaplan‐Meier estimates of the progression‐free interval revealed no significant difference between the group of patients with AR‐poor tumors and patients with AR‐rich tumors. Potential explanations for these results are discussed. The authors conclude that the percentage of AR‐positive nuclei is not a sufficient criterion to predict tumor behavior.