We study the sleep/wake scheduling problem in the context of clustered sensor networks. Unlike most prior work on sleep/wake scheduling that assumed perfect time synchronization in the network, we argue that the synchronization error is non-negligible and demonstrate its effect with a widely used synchronization scheme. We conclude that the design of any sleep/wake scheduling algorithm must take into account the impact of the synchronization error.Our work includes two parts. In the first part, we show that there is an inherent tradeoff between energy consumption and message delivery performance (defined as the message capture probability in this work). We formulate an optimization problem to minimize the expected energy consumption, with the constraint that the message capture probability should be no less than a threshold. In the first part, we assume the threshold is already given. We find the problem to be non-convex, thus cannot be directly solved by conventional convex optimization techniques. However, by investigating the unique structure of the problem, we transform the non-convex problem into a convex equivalent, and solve it using an efficient search method.In the second part, we remove the assumption that the capture probability threshold is already given, and study how to decide it to meet the Quality of Services (QoS) requirement of the application. We observe that in many sensor network applications, a group of sensors collaborate to perform common task(s). Therefore, the QoS is usually not decided by the performance of any individual node, but by the collective performance of all the related nodes. To achieve the collective performance with minimum energy consumption, intuitively we should provide differentiated services for the nodes and favor more important ones. We thus formulate an optimization problem, which aims to set the capture probability threshold for messages from each individual node such that the expected energy consumption is minimized, while the collective performance is guaranteed. The problem turns out to be non-convex and hard to solve exactly. Therefore, we use approximation techniques to obtain a suboptimal solution that approximates the optimum. Simulations show that our approximate solution significantly outperforms a scheme without differentiated treatment of the nodes.