Increasing evidence has supported the use of multiparametric magnetic resonance imaging (mpMRI) for the detection of prostate cancer. However, its role in selecting patients clinically suitable for active surveillance (AS) is still in development. We aimed to find relevant mpMRI features that might be helpful for refinement of the selection of low-risk prostate cancer patients for AS. We also evaluated the interobserver variability in reporting prostate mpMRI features. From 2008 to 2012, 135 patients were selected for AS using Epstein criteria. Baseline mpMRI studies were performed within 3 months of recruitment and reviewed by 2 radiologists who were unaware of the patients' outcomes. The radiologists recorded the mpMRI features using the Prostate Imaging Reporting and Data System (PI-RADS) guidelines. The overall likelihood of the presence of significant prostate cancer was also determined using the Likert and PI-RADS, version 2 (v2), scores. Univariate and multivariate analyses, receiver operating characteristic curves, and Kaplan-Meier survival curves were calculated for the mpMRI features with respect to patient withdrawal from the AS program and failure-free survival (FFS). The interobserver agreement was also evaluated. At a median follow-up time of 31 months (range, 6-80 months), 84 patients (62.2%) were participating in the AS program. In 2 multivariate models, the variables significantly associated with outcomes for both readers were the index lesion size (hazard ratio [HR], 2.34 and 3.13, respectively) and overall PI-RADS, v2, score (HR, 2.51 and 3.21, respectively). The interobserver agreement was higher for the overall Likert and PI-RADS, v2, scores. Overall, the PI-RADS, v2, score and index lesion size were strongly associated with FFS. Overall, the Likert and PI-RADS, v2, scoring systems have been confirmed to be useful, although further improvements are needed.