Background: Accurate risk stratification after radical prostatectomy (RP) is important to help select men at risk of recurrence who will benefit most from postoperative radiation or multi-modal therapy. Increasingly genomic testing is being used in the clinic for this purpose. However, little is known about how these tests predict outcomes in African-American men (AAM), an underserved at risk population. Here we evaluate Decipher within a large Veteran Affairs cohort and compare its performance to the CAPRA-S clinical model for predicting outcomes in patients. Methods: Decipher genomic classifier (GC) scores were generated for 557 patients, who underwent RP at the VA Medical Center Durham between 1989 and 2016. This was a clinically high-risk cohort which all underwent RP. Cox UVA and MVA proportional hazards models and survival c-index were used to compare the performance of Decipher and CAPRA-S for predicting risk of metastasis and PCa specific mortality (PCSM). Results: Overall, 55% (n = 306) of patients in the cohort were AAM. CAPRA-S classified 10.4% as low risk for recurrence while for GC it was 50.4%. With a median follow-up of 9 years, only 40 patients developed metastases and 18 patients died of PCa. In multivariable analyses, both GC (p = 0.044 HR:1.30 95% CI:1.01-1.69) and CAPRA-S (p = 0.037 HR:1.27 95% CI:1.01-1.58) were significant predictors for metastasis within non-AAM; however, only GC (p < 0.001 HR:1.70 95% CI:1.31-2.20), was significant within AAM. GC but not CAPRA-S was a significant predictor of PCSM for both EAM (p = 0.044 HR:1.54 95% CI:1.01-2.53) and AAM (p = 0.002 HR:1.65 95% CI:1.19-2.42). The survival c-index of GC for predicting metastasis 8 years post RP was 0.84 (95% CI: 0.76-0.90) in AAM and 0.70 (95% CI:0.63-0.80) in non-AAM. For PCSM endpoint, it was 0.82 (0.61-0.93) in AAM and 0.73 (95% CI:0.63-0.84) in non-AAM. Conclusions: Our results among non-AAM confirm many prior studies showing that GC is a powerful predictor of metastasis and PCSM. Among AAM, not only was GC a very strong predictor of poor outcome, there was actually a suggestion that GC may perform better among AAM than EAM, though this requires further validation.