BACKGROUND In view of the marked molecular heterogeneity of prostate cancer (PCa), clinical and pathologic parameters alone may be unreliable for predicting disease outcomes after surgical intervention. The development of biomarkers may be helpful to estimate tumor heterogeneity and stratify patients in terms of their risk of progression. Levels of v‐ets avian erythroblastosis virus E26 oncogene homolog (ERG), trefoil factor 3 (TFF3), and serine peptidase inhibitor, Kazal type 1 (SPINK1) are commonly elevated in PCa, but it is unclear whether the evaluation of these 3 markers can help to discriminate patients who will have different clinical outcomes. The authors investigated whether assessment of ERG, TFF3, and SPINK1 expression could help to define clinically relevant, distinct subsets of patients with PCa. METHODS The cohort consisted of 279 men with PCa who underwent radical prostatectomy at Henri Mondor Hospital. Expression levels of ERG, TFF3, and SPINK1 were evaluated immunohistochemically in the prostatectomy specimens. Potential associations of ERG, TFF3, and SPINK1 with age, prostate‐specific antigen (PSA), tumor stage, Gleason score, and biochemical recurrence, defined by PSA failure, were investigated. RESULTS Although prognostic significance was not observed for ERG or TFF3, an exclusive pattern of expression was demonstrated for TFF3 and ERG. SPINK1 expression was observed exclusively in a subgroup of cancers that expressed TFF3 (41 of 175 tumors). Moreover, SPINK1 positivity was identified as predictive of biochemical recurrence in univariate (P = .0009) and multivariate (P = .0003) analyses. CONCLUSIONS The current results suggest that ERG and TFF3 characterize 2 distinct subsets of PCa, with a more aggressive subgroup of TFF3‐expressing tumors that express SPINK1. Together, these findings support a rationale of screening for these biomarkers for prognostic purposes and molecular subtyping of the disease. Cancer 2015;121:1422–1430. © 2015 American Cancer Society. The development of biomarkers may be helpful to estimate tumor heterogeneity and stratify prostate cancer patients in terms of risk of progression. The results from this study suggest that ERG and TFF3 characterize 2 distinct subsets of prostate cancer, including a more aggressive subgroup for TFF3 tumors that express SPINK1, supporting a rationale of screening for these biomarkers for prognostic purposes and molecular subtyping of the disease.