Background: To investigate the prognostic significance of CNA of genes involved in the MMR pathway in localised CaP. Methods: We studied CNA of genes involved in MMR, namely MSH2, MSH3, MSH6, MLH1, PMS2, in 284 patients with intermediate-risk CaP (Toronto cohort), and compared our findings against three public databases (MSKCC and Cambridge cohorts) that included 375 low- to high-risk CaP. The Toronto cohort comprised of 143 and 141 individuals who underwent image-guided radiotherapy (IGRT) and radical prostatectomy (RadP), respectively, while all patients from the public databases underwent RadP. Information on genome-wide copy number alterations (Toronto) was obtained using Affymetrix Oncoscan array. Biochemical relapse-free survival (bRFS) was assessed for clinical outcome. Results: CNA of MSH2, MSH3, MSH6, MLH1, PMS2 were observed in 3.9% (n = 11), 7.7% (n = 22), 3.9% (n = 11), 4.6% (n = 13) and 13.0% (n = 37) of the Toronto cohort, respectively. Distinct patterns of allelic gain and loss were observed for the gene set; gains only for MLH1 and PMS2, and losses only, in all but 1 case, for MSH2, MSH3 and MSH6. In the Toronto cohort, allelic losses of MSH2, MSH3 and MSH6 were determined to be prognostic for poorer bRFS in IGRT patients (HR 2.04, 95% CI 1.01, 4.12, p = 0.048), but not for patients who underwent RadP (HR 1.08, 95% CI 0.49, 2.39, p = 0.84); while gains in MLH1 and PMS2 were not prognostic in either IGRT or RadP patients. A pooled analysis of these genes for all RadP patients from the Toronto and public databases (n = 516) did however indicate that allelic losses of MSH2, MSH3 and MSH6 were significant predictors of poorer bRFS (HR 2.48, 95% CI 1.64-3.77, p < 0.001), but not MLH1 and PMS2 gains. On multi-variable modelling that includes percent genome aberration and pre-treatment PSA levels, allelic losses of MSH2, MSH3 and MSH6remained significant predictors of bRFS for the pooled RadP cohort (HR 1.96, 95% CI 1.27, 3.01, Wald's p < 0.001), but not for IGRT patients (HR 1.50, 95% CI 0.72, 3.12, p = 0.28). Conclusions: We identified a distinct pattern of copy number loss of MSH2, MSH3 and MSH6 genes in localised CaP that appears to be a novel biomarker of failure to definitive treatment.