Background The aim of this study was to determine factors that predict under-evaluation of malignancy in patients diagnosed with atypical ductal hyperplasia (ADH) at ultrasound-guided core needle biopsy (CNB), and to develop a prediction algorithm for scoring the possibility of a diagnosis upgrade to malignancy based on clinical, radiological and pathological factors. Methods The study enrolled patients diagnosed with ADH at ultrasound-guided CNB who subsequently underwent surgical excision of the lesion. Multivariate analysis was used to identify relevant clinical, radiological and pathological factors that may predict malignancy. Results A total of 102 patients with ADH at CNB were identified. Of the 74 patients who underwent subsequent surgical excision, 34 (45.8%) were diagnosed with invasive or in situ malignant foci. Multivariate analysis revealed that age [50 years, microcalcification on mammography, size on imaging [15 mm and a palpable lesion were independent predictors of malignancy. Focal ADH was a negative predictor. A scoring system was developed based on logistic regression models and beta coefficients for each variable. The area under the ROC curve was 0.903 (95% CI: 0.82-0.94), and the negative predictive value was 100% for a score B3.5. Similar findings were observed for a validation dataset of 54 patients at other institutions. Conclusions A scoring system to predict malignancy in patients diagnosed with ADH at CNB was developed based on five factors: age, palpable lesion, microcalcification on mammography, size on imaging and focal ADH. This system was able to identify a subset of patients with lesions likely to be benign, indicating that imaging follow-up rather than surgical excision may be appropriate.