To determine if there is a subgroup of patients with pretreatment PSA > or = 20 ng/ml with a favorable outcome after external beam radiation therapy. We analyzed retrospectively treatment outcomes of 129 patients with pretreatment PSA > or = 20 ng/ml treated in our department from 2/88-8/94. Median patient age was 70 years (range 51-89 years). Tumor stage was T1/T2ab in 68, T2c/T3 in 61 patients. Initial Gleason grade was < 7 in 82 and > or = 7 in 47 patients. Median PSA was 35 ng/ml (mean 45 ng/ml, range 20-191 ng/ml). Ninety-seven patients received four-field conformal external beam radiation therapy. No patient received surgery or hormonal therapy prior to treatment. Median central axis dose was 73 Gy (range 68-79 Gy). Covariates considered in univariate and multivariate analyses included central axis dose, pretreatment PSA, presence of perineural invasion, Gleason score, palpable tumor stage and patient age. bNED failure was defined as a PSA > or = 1.5 and rising on two consecutive determinations. Median follow up was 50 months (range 3-100 months). Overall bNED control for the entire patient population was 22% at five years. Of the covariates analyzed, dose (P < 0.01), stage (P < 0.01), Gleason Score (P < 0.01), and the presence of PNI (P = 0.01) were significant on multivariate analysis. Based on these results, patients could be stratified into two distinct groups. Group I consisted of 19 patients with favorable features including T1/T2ab disease, Gleason Score 2-6, no perineural invasion treated to a dose > 73 Gy to the central axis. Patients in Group II had at least one of the above poor prognostic features or were treated to central axis doses < 73 Gy. The bNED control was significantly higher for patients in Group I than those in Group II (58% vs. 23%, P = 0.0027). There appears to be a favorable subgroup of patients with PSA > or = 20 ng/ml where treating to doses over 73 Gy to the central axis is warranted (four-year bNED rate of 58%). However, because of the small patient numbers, these results will need to be validated with longer follow up.