PURPOSE: The patterns of failure (local and/or regional v metastatic) have been determined for patients with prostate cancer and pretreatment prostate-specific antigen (PSA) levels > or = 20 ng/mL treated with radiation alone with the purpose to design appropriate multimodal treatments. MATERIALS AND METHODS: One hundred twenty patients with pretreatment PSA levels > or = 20 ng/mL were treated with external-beam radiation alone between February 1988 and October 1993. They were arbitrarily divided by PSA levels, 20 to 29.9 ng/mL, 30 to 49.9 ng/mL, and > or = 50 ng/mL, and analyzed in terms of freedom from any failure (no evidence of biochemical disease [bNED], and PSA level < 1.5 ngm/mL and not increasing), as well as freedom from imaging evidence of distant metastasis (fdm). RESULTS: There was no significant difference in short-term outcome by pretreatment PSA level, and thus all patients were pooled for analysis. At 4 years, 81% were fdm and 28% were free of any failure. This suggests that approximately 50% have recurred with local and/or regional disease or undetectable metastatic disease. Multivariate analysis indicated that low palpation stage and higher center of prostate dose were associated with better bNED survival. Multivariate analysis indicated that increasing stage and younger age are significantly associated with increasing distant metastasis. CONCLUSION: Patients with pretreatment PSA levels > or = 20 ng/mL are not optimally treated by irradiation alone. The pattern of failure suggests improvement may come from systemic treatment of metastatic disease and high-dose radiation to improve locoregional disease. To evaluate this, we have begun a multimodal trial of chemohormonal therapy followed by extended-field irradiation.