The role of p53 in the pathogenesis of, and as a predictive biomarker for, localized prostate cancer (PCa) is contested. Recent work has suggested that patterns of p53 nuclear accumulation determined by immunohistochemistry are prognostic, whereas studies using other methods question the role of p53 mutations in predicting outcome. We studied 263 men with localized PCa treated with radical prostatectomy to determine whether p53 nuclear accumulation predicts relapse and disease-specific mortality. We combined two p53 immunohistochemistry scoring systems: (a) percentage of p53-positive tumor nuclei in all major foci of cancer within the prostate; and (b) clustering, where the presence of 12 or more p53-positive cells within a x 200 power field was deemed "cluster positive." Analysis was undertaken using chi2, Kruskal-Wallis, and Mann-Whitney tests for clinicopathological variables and the Kaplan-Meier method, log-rank test, and univariate and multivariate Cox regression modeling for evaluation of contribution to relapse and disease-specific survival. At mean follow-up of 55.1 months (range, 4.9-123.0 months), 39% (102 of 263) of patients had relapsed and 2.3% (6 of 253) had died of PCa. Pretreatment serum prostate-specific antigen concentration, pathological tumor stage, lymph node involvement, Gleason score, and p53 nuclear accumulation, as determined by either percentage score or cluster status, were independent predictors of relapse in multivariate analysis. Clustering of p53-positive cells distinguished between favorable and poor prognosis patients within the lowest p53-positive stratum (>0 to <2%) and was the most discriminatory threshold for predicting relapse in the entire cohort. p53 status predicted outcome in patients with a Gleason score of 5 and above but not those with a score of 4 and below. In patients treated with neoadjuvant hormonal therapy, p53 cluster positivity carried a 90% (19 of 21) risk of relapse by 36 months. All six patients who died from PCa in the period of the study exhibited p53 nuclear accumulation in 20% or more tumor nuclei. This study demonstrates strong relationships between p53 nuclear accumulation and relapse and disease-specific mortality in a large series of localized PCas. Furthermore, the presence of clusters of p53-positive nuclei delineates a group of patients with poor prognosis not identified by traditional scoring methods and supports the hypothesis that p53 dysfunction within PCa may exist in foci of tumor cells that are clonally expanded in metastases.