In this research report, a different and enhanced approach is proposed for mammographic diagnosis for breast biopsy outcome predictions utilizing a neural network classification model and ROC curve evaluation. This classification model contains two components: a training model and a diagnostic model. The training model is based on a two-stage back-propagation neural network approach, along with an iterative training method and an adjustable learning rate. The diagnostic model is able to distinguish and classify malignant breast cancers and benign diseases for breast biopsy outcome predictions. The probability of misclassification error and performance of the proposed classification model were evaluated using the model sensitivity, specificity, and ROC curve analysis.