While wireless sensor networks (WSNs) are typically targeted at large-scale deployment, due to many practical or inevitable reasons, a WSN may not always remain connected. In this paper, we consider the possibility that a WSN may be spatially separated into multiple subnetworks. Data gathering, which is a fundamental mission of WSN, thus may rely on a mobile mule (“mule” for short) to conduct data gathering by visiting each subnetwork. This leads to the problem of minimizing the path length traversed by the mobile mule. We show that minimizing the path length, which may reflect the data gathering latency and the energy consumption of the mule is a generalization of the traveling salesman problem and is NP-complete. Some heuristics based on geometrical properties of node deployment are proposed. Our simulation results show that these heuristics perform very close to optimal solutions in most practical cases.