Image feature extraction was utilized to retrospectively analyze screening mammograms taken prior to the detection of a malignant mass for early detection of breast cancer. The mammograms of 58 biopsy proven breast cancer patients were collected. In each case, the mammograms taken 10 to 18 months prior to cancer detection were evaluated. For each of the two mammographic projections of the abnormal breast, two regions were marked: 1) region one, which corresponded to the site where the malignant mass subsequently developed and 2) a region which appeared similar to region one on the same mammogram. On each projection of the normal breast a third region which corresponds to region one but on the opposite breast was also marked (mirror-image site). Sixty-two texture and photometric image features were then calculated for all of the marked areas. A stepwise discriminant analysis showed that six of these features could be used to best distinguish between the normal and abnormal regions. The best linear classification function resulted in a 72% average classification. At its current stage, the system can be used by a radiologist to examine any pattern in a mammogram. The regions which are flagged by the system have a 72% chance of developing a malignant mass by the time of the next screening. Therefore, further evaluation of these patients (e.g., a screening examination sooner than the normal one year interval) could result in earlier detection of breast cancer. The ultimate goal is to run the system automatically over the whole mammogram and flag any suspicious area.