Background: Distinguishing between early-stage prostate cancer (PCa) and benign prostatic hyperplasia (BPH) has posed a persistent clinical challenge. Exploring the epigenetic characteristics of urinary cell-free DNA (ucfDNA) offers potential for improving diagnosis. In this preliminary study, we aimed to investigate the feasibility of utilizing of urinary cfDNA methylation features as a non-invasive way to detect PCa. Methods: Paired tissue, plasma, and urine samples were collected from 12 PCa patients, along with paired plasma and urine samples from 12 BPH patients, and urine samples from 24 healthy donors. PredicineEPIC, a whole genome-based DNA methylation assay, was implemented to profile methylation features in all these samples. Results: We discovered that the methylation patterns of ucfDNA more closely resembled those of tumor tissue compared to plasma cfDNA, suggesting that urine can serve as an effectual alternative to blood for detecting cancer related signals. We identified 369 differentially methylated regions that contained PCa-specific methylation signals. Subsequently, we employed feature selection and applied a machine learning regression algorithm to construct a classification model. Validation on an independent GEO dataset yielded an impressive AUC of 0.83 for distinguishing between PCa and benign prostatic hyperplasia (BPH). Conclusions: This study demonstrated the potential of ucfDNA-based methylation markers in the early detection of prostate cancer.