Breast cancer is the most prevalent cancer and the main cause of cancer-related death in women worldwide. There are limitations associated with the existing clinical tools for breast cancer detection and alternative modalities for early detection and classification of breast cancer are urgently needed. Here we describe an optical imaging technique, called multispectral diffuse optical tomography (DOT), and demonstrate its ability of non-invasively evaluating nuclear morphometry for differentiating benign from malignant lesions. Photon densities along the surface of the breast were measured to allow for the extraction of three statistical parameters including the size, elongation and density of nuclei inside the breast tissue. The results from 14 patients (4 malignant and 10 benign lesions) show that there exist significant contrasts between the diseased and surrounding normal nuclei and that the recovered nuclear morphological parameters agree well the pathological findings. We found that the nuclei of cancer cells were less-spherical compared with those of surrounding normal cells, while the nuclear density or volume fraction provided the highest contrast among the three statistical parameters recovered. This pilot study demonstrates the potential of multispectral DOT as a cellular imaging method for accurate determination of breast cancer.