Purpose: To evaluate the European Society of Urogenital Radiology (ESUR) score, the Likert scale, tumor contact length (TCL) > 1 cm, and EPE (extraprostatic extension) grade in predicting EPE at multiparametric magnetic resonance imaging (mp-MRI). Methods: Seventy-nine patients who underwent 3-T MRI and were histopathologically confirmed by microblocks were enrolled in this retrospective study. The index lesions were interpreted by two experienced radiologists. Apparent diffusion coefficient (ADC) values were also noted. Weighted κ statistics were used to compare interreader agreement. Univariate logistic regression analysis was performed to define independent predictors of EPE status. Multivariable logistic regression and receiver operating characteristic (ROC) analysis were performed to compare the MRI-based methods and clinical variables (ISUP grade, prostate volume and PSA density) + MRI-based methods for pathologic EPE prediction by using the area under the curve (AUC) value. Results: The mean age was 64.5 years ± 6.2. 33/79 (41.8%) patients had pathologic EPE. As ESUR score showed weak interreader agreement (κ = 0.537), Likert scale, TCL, and EPE grade showed moderate agreement (κ = 0.608, κ = 0.747, κ = 0.647 respectively). Univariate ROC analysis result showed that all MRI-based score systems, mean ADC value, the ISUP grade, prostate volume, PSA density were the best variables in predicting EPE. ROC analysis results of four MRI-based methods showed good diagnostic performance. At multivariate analysis, all clinical models showed excellent diagnostic performance. Conclusion: All four MRI-based methods had good diagnostic performance. Furthermore, consisting of both qualitative and quantitative parameters and being less reader experience dependent, EPE grade was a promising method in predicting EPE. All clinical models showed excellent diagnostic performance.