Purpose: The purpose of this study was to compare the underestimation rate of atypical ductal hyperplasia (ADH) diagnosed by ultrasonography-guided core needle biopsy (US-CNB), ultrasonography-guided vacuum-assisted biopsy (US-VAB) and stereotactic-guided vacuumassisted biopsy (S-VAB), and to evaluate the factors predicting malignancy. Materials and Methods: We retrospectively reviewed the surgical pathology outcomes of 226 lesions that were diagnosed with atypical ductal hyperplasia on US-CNB, US-VAB, and S-VAB between January 2005 and November 2015. Clinical and radiological features including age, size on imaging, family history, imaging features, residual lesion, symptom, Brest Imaging Reporting and Data System (BI-RADS) category, synchronous cancer were recorded and compared according to the biopsy methods. The underestimation rate of ADH was calculated and the factors were analyzed associated with underestimation of ADH. Results: Of 226 atypical ductal hyperplasia cases, 65 cases proved to be ductal carcinoma in situ and 21 cases proved to be invasive cancer on surgical excision. The underestimation rate of ADH on US-CNB, US-VAB and S-VAB were 48.25% (69/143), 21.62% (8/37), and 19.57% (9/46). Among 83 ADH cases on VAB, cases without residual lesion on imaging (n=35) is associated with the low underestimation rate (5.7% [2/35] vs. 31.3% [15/48], p=0.004). Significant predictive factors for malignancy were size on imaging ≥1 cm (P=0.003), mass or mass with calcification on imaging (P=0.029), nipple discharge or palpable lesion (P<0.001), and higher BI-RADS category (P<0.001). There was no significant difference in age, family history and synchronous cancer. Conclusion: The underestimation rate of ADH on VAB was lower than that of US-CNB. Lesion size on imaging ≥1 cm, mass forming lesion, nipple discharge or palpable lesion, higher BIRADS category lesion, and residual lesion were likely to be malignant and predictive factor of malignancy.