We aimed to evaluate the diagnostic performance of thyroid ultrasonography (US) in each category of the Bethesda system and analyze false positive/negative findings using US. This retrospective study included 622 thyroid nodules in 592 patients. The sensitivity, specificity, negative predictive value (NPV), positive predictive value (PPV) and accuracy of US in each category of the Bethesda system were evaluated. False positive/negative cases of US were analyzed. Out of the 622 total thyroid FNAs, 179 (28.8%) were malignant. The malignancy rates for the 6 categories were as follows: I (nondiagnostic): 9.7%, II (benign): 2.5%, III (atypia/follicular lesion of undetermined significance): 37.5%, IV (suspicious for follicular neoplasm): 5.7%, V (suspicious for malignancy): 100%, and VI (malignancy): 100%. The accuracies of US for the 6 categories were 92.5%, 95.6%, 70.8%, 94.3%, 95%, and 92.4% in category order. US showed the lowest sensitivity (50%) in Category IV. Category III demonstrated relatively low sensitivity (66.7%) and specificity (73.3%) due to a high incidence of follicular variant of papillary thyroid carcinoma and a low number of category III nodules. The most optimal performance of US was revealed in Category I with 88.9% sensitivity and 92.9% specificity. In 22 US false positive cases, the most frequent finding was associated with marked hypoechogenicity and the least finding was noncircumscribed margin. The most common US features of 19 false negative cases were circumscribed iso or hypoechoic nodules. These results highlight the excellent diagnostic performance of US in category I of the Bethesda system and the lowest sensitivity of US in category IV. Awareness of US interpreters regarding these pitfalls can minimize false positive/negative diagnoses and prevent unnecessary interventions.