OBJECTIVE Frozen-section evaluation of ovarian tumors can be used to establish a histopathologic diagnosis and guide the surgeon to perform the appropriate surgical procedure. A retrospective study was conducted to determine the accuracy of frozen-section diagnosis of ovarian tumors. METHODS Frozen- and permanent-section diagnoses were divided into three categories (benign, borderline, and malignant). The sensitivity, specificity and predictive values, and 95% percent confidence intervals of each frozen-section diagnosis were determined. RESULTS Three hundred eighty-three ovarian tumors that underwent frozen-section evaluation between June 1983 and June 1993 were studied. The final histopathologic diagnosis was 61.1% benign, 7.6% borderline, and 31.3% malignant. Frozen section was accurate in 92.7% of all cases and inaccurate in 7.3%. The sensitivity for malignant tumors was 92.5% tumors (95% confidence intervals 87.7% to 97.2%), the sensitivity for borderline tumors was 44.8% (95% confidence interval 26.4% to 63.2%). The specificity for benign tumors was 92.0% (95% confidence interval 88.6% to 95.4%) but increased to 97.9% (95% confidence interval 96.1% to 99.7%) if borderline tumors were excluded. The positive predictive value and 95% confidence intervals were 92.0% (88.6% to 95.4%) for benign tumors, 65% (43.6% to 86.5%) for borderline tumors, and 99.1% (97.3% to 100.0%) for malignant tumors. Thirteen of 16 (81%) ovarian lymphomas and tumors metastatic to the ovary were correctly identified by intraoperative frozen section. The sensitivity for borderline serous tumors was 64.3% and for borderline mucinous tumors 30.8% (p = 0.48). CONCLUSIONS With the exception of borderline tumors, the sensitivity and specificity of frozen-section diagnosis of ovarian tumors are high. Borderline tumors remain difficult to accurately diagnose at frozen section because of the extensive sampling required. Frozen-section diagnoses have important implications regarding the type and extent of surgery performed at the initial operation.