Breast cancer is the most commonly diagnosed cancer for women. It has a higher death rate than any other types of cancers for women in the United States. Detecting tumors in their early stages through screening examinations is key to reducing breast cancer mortality. Mammography has its limitations and might miss breast cancers in some cases. The high false positives rate of screening ultrasound results in additional imaging/biopsy and anxiety. In this work, we present an effective method to automatically classify breast cancer images using convolutional neural network (CNN). CNN is inspired by connectivity patterns between biological neurons. This is an effective method of image classification in supervised machine learning because the network has ability to learn the parameters that in traditional approaches are hand-engineered. The classification procedure usually includes data augmentation, training the network, and testing the network. We initially obtained 84.3% overall accuracy on breast cancer mammography classifications, even though there exists some unexpected problems such as the low precision accuracy. For the future work, we plan to improve the algorithm in order to make the CNN fully learn this classification task. Besides this, In order to speed up CNN training, we will explore different toolboxes to implement CNN such as MXNet. Ultimately, we will apply this method on breast cancer ultrasound imaging.