Temporal feature binding in vision refers to the process by which features of objects presented one after the other at the same spatial location are correctly bound together. In this paper, we describe a computational model of putative neural mechanisms that would produce this behaviour. These simulations highlight the role of transient attentional enhancement in mediating the temporal binding of features into working memory. This model builds upon previous approaches, and explains a range of behavioural findings relating to the patterns of illusory conjunctions observed in experiments. Further, it provides a parsimonious account of a counter-intuitive pattern of reaction time data.