To determine whether multiple features of immunohistochemical staining of the androgen receptor (AR) in prostate cancer could reliably predict androgen dependence, tumor biopsy specimens from 30 patients (stages A-D2) were stained using anti-peptide antibodies to the amino- and carboxyl-terminal of the AR. Measurements were made of the mean area and total amount (i.e., integrated optical density) of AR staining in at least 20 fields per section using a color video image analysis system, and the mean intensity of AR staining per cell and the percentage of AR positive tumor cells were derived. Video image analysis measurement identified quantitative differences in AR staining between the two antibodies, suggesting that this approach may provide a means of identifying receptor variants in prostate tumors. The AR staining measurements were analyzed by discriminant function analysis to assign individual cases to good and poor clinical outcome groups. AR staining features measured with a single antibody (e.g., amino-terminal) were sufficient to predict outcome following hormonal therapy in stage D2 patients (predictive value, 1.0), whereas all features of AR staining measured with both antibodies were required for the entire patient group (predictive value, 0.97). The principal discriminant in both patient groups contributing to the correct assignment of outcome was the mean intensity of AR staining per cell. These findings suggest that AR staining features measured by video image analysis have the potential to predict outcome in prostate cancer.