Background: To assess the accuracy of ultrasound in differentiating endometrioma from ovarian cancer and to describe pattern recognition for atypical endometriomas mimicking ovarian cancers. Materials and Methods: Patients scheduled for elective surgery for adnexal masses were sonographically evaluated for endometrioma within 24 hours of surgery. All examinations were performed by the same experienced sonographer, who had no any information of the patients, to differentiate between endometriomas and non-endometriomas using a simple rule (classic ground-glass appearance) and subjective impression (pattern recognition). The final diagnosis as a gold standard relied on either pathological or post-operative findings. Results: Of 638 patients available for analysis, 146 were proven to be endometriomas. Of them, the simple rule and subjective impression could sonographically detect endometriomas with sensitivities of 64.4% (94/146) and 89.7% (131/146), respectively. Of 52 endometriomas with false negative tests by the simple rule, 13 were predicted as benign masses and 39 were mistaken for malignancy. Solid masses and papillary projections were the most common forms mimicking ovarian cancer, consisting of 38.5% of the missed diagnoses. However, with pattern recognition (subjective impression), 32 from 39 cases mimicking ovarian cancer were correctly predicted for endometriomas. All endometriomas subjectively predicted for ovarian malignancy were associated with high vascularization in the solid masses. Conclusions: Pattern recognition of endometriomas by subjective assessment had a higher sensitivity than the simple rule in characterization of endometriomas. Most endometriomas mimicking ovarian malignancy could be correctly predicted by subjective impression based on familiarity of pattern recognition.