While recent person re-identification (ReID) methods achieve high accuracy in\na supervised setting, their generalization to an unlabelled domain is still an\nopen problem. In this paper, we introduce a novel unsupervised disentanglement\ngenerative adversarial network (UD-GAN) to address the domain adaptation issue\nof supervised person ReID. Our framework jointly trains a ReID network for\ndiscriminative features extraction in a source labelled domain using identity\nannotation, and adapts the ReID model to an unlabelled target domain by\nlearning disentangled latent representations on the domain. Identity-unrelated\nfeatures in the target domain are distilled from the latent features. As a\nresult, the ReID features better encompass the identity of a person in the\nunsupervised domain. We conducted experiments on the Market1501, DukeMTMC and\nMSMT17 datasets. Results show that the unsupervised domain adaptation problem\nin ReID is very challenging. Nevertheless, our method shows improvement in half\nof the domain transfers and achieve state-of-the-art performance for one of\nthem.\n