Deep unsupervised representation learning has recently led to new approaches\nin the field of Unsupervised Anomaly Detection (UAD) in brain MRI. The main\nprinciple behind these works is to learn a model of normal anatomy by learning\nto compress and recover healthy data. This allows to spot abnormal structures\nfrom erroneous recoveries of compressed, potentially anomalous samples. The\nconcept is of great interest to the medical image analysis community as it i)\nrelieves from the need of vast amounts of manually segmented training data---a\nnecessity for and pitfall of current supervised Deep Learning---and ii)\ntheoretically allows to detect arbitrary, even rare pathologies which\nsupervised approaches might fail to find. To date, the experimental design of\nmost works hinders a valid comparison, because i) they are evaluated against\ndifferent datasets and different pathologies, ii) use different image\nresolutions and iii) different model architectures with varying complexity. The\nintent of this work is to establish comparability among recent methods by\nutilizing a single architecture, a single resolution and the same dataset(s).\nBesides providing a ranking of the methods, we also try to answer questions\nlike i) how many healthy training subjects are needed to model normality and\nii) if the reviewed approaches are also sensitive to domain shift. Further, we\nidentify open challenges and provide suggestions for future community efforts\nand research directions.\n