Histology images are inherently symmetric under rotation, where each\norientation is equally as likely to appear. However, this rotational symmetry\nis not widely utilised as prior knowledge in modern Convolutional Neural\nNetworks (CNNs), resulting in data hungry models that learn independent\nfeatures at each orientation. Allowing CNNs to be rotation-equivariant removes\nthe necessity to learn this set of transformations from the data and instead\nfrees up model capacity, allowing more discriminative features to be learned.\nThis reduction in the number of required parameters also reduces the risk of\noverfitting. In this paper, we propose Dense Steerable Filter CNNs (DSF-CNNs)\nthat use group convolutions with multiple rotated copies of each filter in a\ndensely connected framework. Each filter is defined as a linear combination of\nsteerable basis filters, enabling exact rotation and decreasing the number of\ntrainable parameters compared to standard filters. We also provide the first\nin-depth comparison of different rotation-equivariant CNNs for histology image\nanalysis and demonstrate the advantage of encoding rotational symmetry into\nmodern architectures. We show that DSF-CNNs achieve state-of-the-art\nperformance, with significantly fewer parameters, when applied to three\ndifferent tasks in the area of computational pathology: breast tumour\nclassification, colon gland segmentation and multi-tissue nuclear segmentation.\n