This study evaluated how to manage Bethesda category III (Bethesda III) (atypia of undetermined significance/follicular lesion of undetermined significance [AUS/FLUS]) thyroid nodules according to the Korean Thyroid Imaging Reporting and Data System (K-TIRADS) to reduce unnecessary surgeries. A total of 161 thyroid nodules diagnosed as Bethesda III underwent surgery from 2016 to 2019. Ultrasonography-guided fine-needle aspiration (US-FNA) or core needle biopsy (CNB) was used for repeat examination. K-TIRADS category was assigned to the thyroid nodules. The proportion of malignancy in Bethesda III nodules confirmed by surgery were significantly increased in proportion relative to K-TIRADS with 60.0% low suspicion, 88.2% intermediate suspicion, and 100% high suspicion nodules (p < 0.001). The proportion of malignancy in AUS and FLUS were significantly different (94.2% vs. 40.0% p = 0.003). The proportion of malignancy in AUS increased with K-TIRADS categories, but there was no difference in FLUS. All K-TIRADS high suspicion nodules were AUS as papillary carcinomas (99%), while 80% of FLUS nodules and 50% of follicular carcinomas showed K-TIRADS low suspicion. In 116 nodules with repeat FNA or CNB after initial Bethesda III results, the conclusive result rate was significantly increased in proportion to K-TIRADS with 58.3% low suspicion, 83.3% intermediate suspicion, and 88.8% high suspicion nodules (p = 0.015). K-TIRADS low suspicion nodules of Bethesda III nodules should be managed after risk-benefit consideration rather than immediate surgery or repeat examination. K-TIRADS for Bethesda III nodules can predict papillary carcinoma well, but not follicular carcinoma.