Frozen section (FS) diagnosis was routinely performed in a large series of nonpalpable breast lesions from 1977 through 1991. The original FS diagnoses of 672 patients were classified in four categories (1 = benign lesion, 2 = in situ carcinoma, 3 = invasive carcinoma, 4 = deferred diagnosis) and compared with the diagnoses obtained at review of the permanent paraffin sections to estimate the accuracy of FS. A review of the mammographic pattern of the lesion was also performed. Frozen section diagnostic conclusion was deferred to permanent paraffin sections in only 22 cases (3.3%). Benign or malignant (grouping in situ and invasive carcinomas) FS diagnoses were accurate in 623 of 650 cases (95.8%). Overall, the prevalence of malignant lesions was 44.8% with a benign/malignant ratio of 1.2. The diagnosis was modified on the basis of permanent sections in 27 cases (4.2%) with three false positives and 24 false negatives. Sensitivity and specificity of FS diagnoses were 91.7 and 99.2%, respectively. When the comparison between FS and histologic diagnoses was analyzed according to the mammographic pattern, sensitivity among patients with microcalcifications as the only alteration was lower (88.8%) than among patients with opacities (94.9%). On the basis of these results, FS is to be considered a feasible and reliable diagnostic procedure in nonpalpable breast lesions, particularly in cases excised because of a mammographic opacity that is identifiable on gross examination of the surgical specimen.