Background: Biochemical recurrence (BCR) of prostate cancer (PCA) patient indicates poor prognosis and more complex treatment. However, the tools for predicting BCR remain controversial. This study performed a comprehensive analysis of gene expression and DNA methylation data to establish a BCR classifier for PCA. Methods: We analyzed DNA methylation data and mRNA expression datasets for over 1000 clinical samples from the TCGA cohort. We integrated RNA-sequencing and DNA methylation data to identify DNA methylation (DNAm) driven differentially expressed genes (DEGs). Then, these genes were determined to build a prognostic classifier by Kaplan-Meir analysis and 1000 times Cox LASSO regression with 10-fold CV. Furthermore, other datasets were utilized to validate externally. Findings: We identified a total of 65 DNA methylation-driven DEGs, and developed a four-DNAm-driven DEGs-based classifier which has been shown to accurately predict the BCR-free survival of patients in TCGA and validation sets. Additionally, multivariate Cox regression analysis determined the Classifier as an independent predictor of BCR (hazard ratio [HR] 0.426 [95% CI 0.194–0.936], P < 0.05). Then, a nomogram based on the Classifier and Gleason score was constructed, which show a predictive accuracy significantly higher than that of each variable alone (AUC at 1-, 3-, 5 years 0.701,0.799, 0.838). Interpretation: We identified and validated the BCR classifier of PCA constructing by four DEGs, suggesting this Classifier has the potential to guide treatment decisions for patients at differing risks of BCR. Funding Statement: This study was funded by the National Natural Science Foundation of China. Declaration of Interests: The authors declare no other competing interests.