Working memory (WM) is a cognitive system that holds and manipulates information over short periods. The process of updating allows WM to selectively keep track of task-relevant information so that ongoing mental activity can proceed smoothly. The thesis established the first fine-grained temporal analysis of the cognitive demands associated with WM updating, while also presenting recall behaviour and self-reported strategy data. Chapter 1 provided a theoretical background and reviewed experimental investigations of WM updating with a particular emphasis on two updating tasks, running span and n-back. Chapter 2 then described an experiment in which the recruitment of cognitive resources was charted while participants completed a running span task. The chapter demonstrated that the resource demands associated with running span follow a specific time course that is absent in serial recall tasks that do not require WM updating. Chapter 3 examined whether the updating process profiled in the previous experiment is sensitive to the temporal parameters of the task. By tracking resource demands during running span with different rates of item presentation, the chapter showed that updating is time-consuming and thus limited to the slow-paced task. Chapter 4 presented a strategy induction procedure in which participants were instructed to perform running span using active updating or passive listening. This chapter compared the behavioural impact of the two strategies and replicated the time course of updating observed in the first two experiments using converging operations. Chapter 5 extended the investigation to another updating task and showed that the updating-related demands in n-back resembled those in running span. In addition, memory performance in terms of both serial position curves and recall/recognition errors was examined in each respective chapter. Self-report data were also presented in Chapters 3-5 to further understand the cognitive mechanisms of WM updating. Finally, Chapter 6 brought together findings from the four experiments and discussed them in the context of possible computational accounts of WM updating. It also considered the limitations of the present work and discussed possible avenues for further research.