BACKGROUND.: The Thyroid Imaging Reporting and Data System (TI-RADS) was proposed based on a scheme similar to Breast Imaging Reporting and Data System (BI-RADS) lexicon used in breast lesions. The purpose of this study was to evaluate its interobserver variability and accuracy. METHODS.: We included 498 nodules in 437 patients undergoing thyroidectomy. Two endocrine surgeons and 2 endocrinologists independently reviewed sonographic images. RESULTS.: There was moderate to substantial interobserver agreement for final assessment category (kappa = 0.61). The overall sensitivity, specificity, and negative predictive value (NPV) were 94%, 43%, and 96%, respectively. Positive predictive values (PPVs) for categories 4 and 5 were 32% and 60%. The sensitivity was 92%, 99%, 96%, and 89%, whereas the specificity was 25%, 37%, 41%, and 62% for tumor sizes of <2, 2 to 3, 3 to 4, and >4 cm, respectively. CONCLUSION.: TI-RADS is a helpful but not optimal reporting tool in characterizing thyroid lesions. Tumor size has a considerable impact on interobserver concordance and diagnostic performance.