Clinical management of ductal carcinoma in situ (DCIS) remains a challenge because significant proportions of patients experience recurrence after conservative surgical treatment. Unfortunately, it is difficult to prospectively identify, using objective criteria, patients who are at high risk of recurrence and might benefit from additional treatment. We conducted a multi-institutional, collaborative case-control study to identify nuclear morphometric features that would be useful for identifying women with DCIS at the highest risk of recurrence. Tissue sections of archival breast tissue of 29 women with recurrent and 73 matched women with nonrecurrent DCIS were stained for DNA, and nuclei in the DCIS lesions were evaluated by image analysis. A clear correlation between mean fractal2_area (FA2) and nuclear grade was observed (P < 0.001), allowing an objective determination of nuclear grade. Several nuclear morphometric features, including mean and variance of variation of radius, mean area, mean and variance of frequency of high boundary harmonics (FQH), and variance in sphericity, were found to be useful in discriminating recurrent from nonrecurrent DCIS subjects. However, the nuclear features associated with recurrence differed between high- and low-grade lesions. For lesions with high FA2 (nuclear grade 3), mean variation of radius, mean FQH, and mean area alone yielded recurrence odds ratios of 4.55 [95% confidence interval (CI) 0.45-45.96], 3.86 (95% CI, 0.88-16.98), 2.90 (95% CI, 0.31-27.2), respectively. Using a summed feature model, high-FA2 lesions showing three poor prognostic features had an odds ratio of 15.63 (95% CI, 1.22-200), compared with those with zero or one poor prognostic feature. Lesions with low mean FA2 (nuclear grade 1 or 2) showing high variances in sphericity and FQH had an odds ratio of 7.71 (95% CI, 1.77-33.60). Addition of other features did not enhance the odds ratio or its significance. These results suggest that nuclear image analysis of DCIS lesions may provide an adjunctive tool to conventional pathological analysis, both for the objective assessment of nuclear grade and for the identification of features that predict patient outcome.