Computer-aided breast cancer detection is at the core of this thesis. Breast cancer is a prominent cause that harvests the lives of women. Early detection is of utmost importance to combat time, which is vital when the test results come in. With the help of deep learning, binary classification can be achieved with different CNNs to aid medical staff, and the experiment will help find the most prominent one(s). The CNNs explored in this thesis consist of; Inception-ResNet-v2, Xception, ResNet50, DenseNet201,Inception-v3 and VGG16. Each of the models underwent transfer learning, which allowed a shorter training period. The data collected for the experiment is from the BreakHis database, consisting of histopathological imagery of various magnification levels (40X, 100X, 200X, and 400X). The experiment started with data augmentation, which allowed the CNN models to be trained more effectively and aid class balancing. This was met by increasing the training data by manipulating the data from the source. This thesis concludes that Inception-ResNet-v2, Xception, ResNet50, DenseNet-201, and Inception-v3share similar but not identical performance metrics. The VGG16 was the only model that was not on par with the others. The magnification levels caused differentiated results for each of the models. DenseNet-201 had the best accuracy in the 40X magnification category (99.5).