The assumption that tumor grading is best performed in the region having the least differentiation was tested in a series of mammary ductal carcinomas by quantitating the relative amounts of neoplastic cells that remain within ducts (intraductal tumor [IN]) or participate in glandular formations (differentiated infiltrating tumor [DI]). Field-to-field variation in these two elements displayed a linear correlation with their overall level in each tumor. Since having less than 10% DI + IN is a marker for recurrence among breast cancer patients, this marker was applied to find variant regions of potential biological importance. Such regions were common, but their distribution was random, and they varied so much within themselves that they were not significantly different from the remainder of the tumor. Furthermore, the existence of such poorly differentiated regions did not correlate with a poor prognosis. Grading should be performed on a sample that is representative of the entire tumor, rather than its least differentiated part.