The paper presents a method for mitosis detection in breast cancer histology images using deep max-pooling convolutional neural networks (DNNs). The DNNs are trained to classify each pixel in the images, using a patch centered on the pixel as context. The approach won the ICPR 2012 mitosis detection competition, outperforming other contestants. The method is conceptually simple, involving a DNN that operates directly on raw RGB data, and it is tested on a publicly available dataset. The DNN is trained to differentiate patches with a mitotic nucleus close to the center from all other windows. Mitosis in unseen images are detected by applying the classifier on a sliding window and post-processing the outputs. The approach is rotationally invariant and can handle the challenge of mitotic nuclei appearing similar to non-mitotic structures. The performance of the method is evaluated using the MITOS dataset, and it achieves an F-score of 0.782, significantly higher than the closest competitor. Future work aims to validate the approach on larger datasets and compare its performance with expert histologists.The paper presents a method for mitosis detection in breast cancer histology images using deep max-pooling convolutional neural networks (DNNs). The DNNs are trained to classify each pixel in the images, using a patch centered on the pixel as context. The approach won the ICPR 2012 mitosis detection competition, outperforming other contestants. The method is conceptually simple, involving a DNN that operates directly on raw RGB data, and it is tested on a publicly available dataset. The DNN is trained to differentiate patches with a mitotic nucleus close to the center from all other windows. Mitosis in unseen images are detected by applying the classifier on a sliding window and post-processing the outputs. The approach is rotationally invariant and can handle the challenge of mitotic nuclei appearing similar to non-mitotic structures. The performance of the method is evaluated using the MITOS dataset, and it achieves an F-score of 0.782, significantly higher than the closest competitor. Future work aims to validate the approach on larger datasets and compare its performance with expert histologists.