A retrospective study of 321 patients who underwent localizing mammography and excisional biopsy of the breast from 1984 to 1985 was performed. The study was undertaken to refine selection criteria for biopsy in women with nonpalpable mammographic abnormalities by comparing mammographic features and impression with histologic findings. Twenty-eight of 36 (78 per cent) noninfiltrating carcinomas presented with microcalcifications alone; in contrast, 27 of 39 (69 per cent) infiltrating carcinomas presented with a mass alone. As the number of microcalcifications increased, so did the incidence of carcinoma. The size of the mass was not a guide for predicting carcinoma. Although only 11 of 75 carcinomas presented as a mass with microcalcifications, 11 of 21 calcified masses were carcinoma. There were no significant differences in the mammographic presentation between ductal and lobular carcinoma. The sensitivity of the mammographic impression was 48/75 (0.64), and the specificity was 221/246 (0.898). The false-positive rate was 25/73 (0.34), and the false-negative rate was 10/141 (0.07). From this study, we concluded 1, the incidence of noninfiltrating carcinoma was significantly higher and the incidence of positive nodes was significantly lower in nonpalpable abnormalities than in palpable masses; 2, noninfiltrating carcinomas were generally associated with microcalcifications alone, but infiltrating carcinomas were generally associated with a mass alone, and 3, the diagnostic accuracy of mammography was limited by under-interpretation of the subtler signs of noninfiltrating carcinoma and by over-interpretation of mammographic findings generally accepted as criteria for carcinoma.