Mass screening using mammography is, at present, the only viable and effective method to detect breast cancer. It is difficult, however, to distinguish between benign and malignant microcalcifications associated with breast cancer. This difficulty results in a significant increase in the number of biopsy examinations. Most of the minimal breast cancers are currently detected by the presence of micro-calcifications. the major problems are the relatively low-positive predictive value and the high false-positive rate necessary to maximize sensitivity for minimal breast cancer detection. It is the long-term (Phase I and Phase II) objective of this project to be able to reduce the false-positive rate of breast cancer detection, while maintaining high specificity. The objective of Phase I is to develop a computerized artificial neural network-based mammogram analysis system. Basic steps proposed are feature selection for the microcalcification regions in the mammograms, designing and training the neural network, and testing and verification of classification accuracy of neural network algorithms. Image structure features will be selected from a set of mammograms with benign and malignant microcalcifications to provide good discrimination between them. The proposed system will possess the ability to accurately segment such regions. This system can be subsequently refined to provide high specificity.