Objective: To determine whether the categories defined in the Breast Imaging Reporting and Data System (BI-RADS) are useful predictors of malignancy. Methods: A total of 348 cases with benign and malignant breast diseases were collected. Mammographic and pathologic findings were reviewed. Mammograms of 348 cases were characterized according to BI-RADS descriptors and were categorized by the final assessment categories. Results: Of the all 348 patients, 232 (67%) were carcinomas. Significantly more masses with calcification and speculation were found in breast cancers than in benign breast diseases. BI-RADS final assessment categories were category 2 and 3 in 106 cases, in which 75% (79/106) were benign; category 4 and 5 in 242 cases, in which 85% (205/242) were carcinomas. BI-RADS categories 4 and 5 are useful predictors of relative likelihood of malignancy. The features with higher positive predictive values for carcinomas were irregular shape, indistinct or speculated margins, fine or linear calcification morphology, and regional calcification distribution. Conclusion: BI-RADS lexicon is successful in providing a standardized language for physicians to describe lesion morphology. BI-RADS category is useful for predicting the presence of malignancy.