QoS-Aware web services composition has recently received much attention. While most work focused on service selection, we study QoS in another stage of the life cycle of composite services, namely, scheduling. An interesting problem is whether we can obtain better QoS via scheduling even when the component services have been fixed. In this paper, we propose an approach to find an optimal (near-optimal) schedule with the least cancellation cost, which can further improve the overall QoS of composite services. An approach to analyze the expected cancellation cost of a schedule of a composite service is proposed and QoS-Aware service scheduling is formalized as a Constraint Satisfaction Optimization Problem (CoSOP). Two algorithms - heuristic backtracking and genetic algorithm - are presented to find an optimal (near-optimal) schedule, and their performance is studied by simulations. Preliminary experimental results show that our approach is effective.