Objective: The aim of this study was to describe the ultrasound features of benign and malignant thyroid nodules and evaluate the likelihood of malignancy associated with each feature according to the Bethesda System for Reporting Thyroid Cytopathology and histopathology. With this analysis, we propose a new TI-RADS classification system. Materials and methods: The likelihood of malignancy from ultrasound features were assessed in 1413 thyroid nodules according to the Bethesda System for Reporting Thyroid Cytopathology and histopathological findings. A score was established by attributing different weights to each ultrasound feature evaluated. Results: Features positively associated with malignancy in bivariate analysis received a score weight of +1. We attributed a weight of +2 to features which were independently associated with malignancy in a multivariate analysis and +3 for those associated with the highest odds ratio for malignancy (> 10.0). Hence, hypoechogenicity (graded as mild, moderate or marked, according to a comparison with the overlying strap muscle), microcalcification and irregular/microlobulated margin received the highest weights in our scoring system. Features that were negatively associated with malignancy received weights of -2 or -1. In the proposed system a cutoff score of 2 (sensitivity 97.4% and specificity 51.6%) was adopted as a transition between probably benign (TI-RADS 3) and TI-RADS 4a nodules. Overall, the frequency of malignancy in thyroid nodules according to the categories was 1.0% for TI-RADS 3, 7.8% for TI-RADS 4a, 35.3% for TI-RADS 4b, and 84.7% for TI-RADS 5. Conclusion: A newly proposed TI-RADS classification adequately assessed the likelihood of malignancy in thyroid nodules. Arch Endocrinol Metab. 2017;61(3):211-21.