Automatic segmentation of hippocampus in MR images is a vital step towards the development of computer-aided diagnosis systems. Owing to the quite ambiguous edges of hippocampus, recent deep learning approaches applied on hippocampus segmentation, without attaching importance to the edge information, have a restriction on producing accurate edges. This paper presents a method to automatically segment hippocampus using a novel edge-aware fully convolutional network (FCN) ending with a dense conditional random field (CRF) layer. This method achieves a more precise edge segmentation by incorporation edge information into the loss function. Validation results on the ANDI dataset and NITRC dataset show that the proposed method produces hippocampus segmentation results of high quality, scores an average dice similarity coefficient up to 87.31% and performs better than the state-of-the-art approaches. Our method may contribute to clinical diagnosis of probable Alzheimer’s Disease.