Several studies have shown an association between high nuclear grade or necrosis of ductal carcinoma in situ (DCIS) lesions and the risk of local disease recurrence in patients with DCIS treated surgically with less than mastectomy. Although criteria for separating low from high nuclear grade lesions have been published, no information exists regarding interobserver reproducibility (IR). To assess IR in the classification of DCIS, six surgical pathologists from four institutions used the Lagios grading system to grade 125 DCIS lesions. Before meeting to evaluate the cases, a training set of 12 glass slides, including cases chosen to present conflicting cues for classification, was mailed to the participants with a written criteria summary. This was followed by a working session in which criteria were reviewed and agreed on. The pathologists then graded the lesions independently. The area of interest was marked on each slide before grading. After initial grading, the pathologists met again to resolve discrepant lesion classifications. A complete agreement among raters was achieved in 43 (35%) cases, with five of six raters agreeing in another 45 (36%) cases. In no case did two raters differ by more than one grade. The pairwise kappa agreement values ranged from fair to substantial (0.30 to 0.61). Generalized kappa value indicated moderate agreement (0.46, standard error = 0.02). Kappa statistics for the distinction between grades 1 and 2 and 2 and 3 were 0.29 and 0.48, respectively, (standard error = 0.02). Only one of the six raters differed significantly in scoring. With adherence to specific criteria, IR in the classification of DCIS cases can be obtained in most cases. Although these pathologists made a few grading system modifications, further refinements are needed, especially if grading will influence future therapy.