Background: The low specificity of prostate-specific antigen (PSA) has resulted in the overdiagnosis and overtreatment of clinically indolent prostate cancer (PCa). We aimed to identify a urine exosomal circular RNA (circRNA) classifier that could detect high-grade (Gleason score [GS]7 or greater) PCa. Methods: We did a three-stage study that enrolled eligible participants, including PCa-free men, 45 years or older, scheduled for an initial prostate biopsy due to suspicious digital rectal examination findings and/or PSA levels (limit range, 2.0-20.0 ng/mL), from four hospitals in China. We used RNA sequencing and digital droplet polymerase chain reaction to identify 18 candidate urine exosomal circRNAs that were increased in 11 patients with high-grade PCa compared with 11 case-matched patients with benign prostatic hyperplasia. Using a training cohort of eligible participants, we built a urine exosomal circRNA classifier (Ccirc) to detect high-grade PCa. We then evaluated the classifier in discrimination of GS7 or greater from GS6 and benign disease on initial biopsy in two independent cohorts. We used the sensitivity, specificity, and area under the receiver operating characteristic curve (AUC) to evaluate diagnostic performance, and compared Ccirc with standard of care (SOC) (ie, PSA level, age, race, and family history). Results: Between June 1, 2016, and July 31, 2019, we recruited 356 participants to the training cohort, and 442 and 325 participants to the two independent validation cohorts. We identified a Ccirc containing five differentially expressed circRNAs (circ_0049335, circ_0056536, circ_0004028, circ_0008475, and circ_0126027) that could detect high-grade PCa. Ccirc showed higher accuracy than SOC to distinguish individuals with high-grade PCa from controls in both the training cohort and the validation cohorts. (AUC 0.831 [95% CI 0.765-0.883] vs 0.724 [0.705-0.852], P = 0.032 in the training cohort; 0.823 [0.762-0.871] vs 0.706 [0.649-0.762], P = 0.007 in validation cohort 1; and 0.878 [0.802-0.943] vs 0.785 [0.701-0.890], P = 0.021 for validation cohort 2). In all three cohorts, Ccirc had higher sensitivity (range 71.6-87.2%) and specificity (82.3-90.7%) than did SOC (sensitivity, 42.3-68.2%; specificity, 40.1-62.3%) to detect high-grade PCa. Using a predefined cut point, 202 of 767 (26.3%) biopsies would have been avoided, missing only 6% of patients with dominant pattern 4 high-risk GS 7 disease. Conclusions: Ccirc is a potential biomarker for high-grade PCa among suspicious men.