Abstract: In the traditional single-sink sensor network, query dissemination and data collection are based on a fixed infrastructure. This infrastructure has some disadvantages, such as consuming energy of the nodes on the key path too quickly, the singleness of routing algorithm, the invalidation of the sink node, etc. To solve these problems, the research of multi-sink sensor networks is deployed. The system architecture of the multi-sink sensor network is proposed and a topology discovering and maintaining policy is provided. the system architecture includes task manager, proxy nodes, sink nodes and common nodes. Then a routing algorithm based on minimum energy consumption is provided. Because the minimum energy consumption routing algorithm consumes energy of the nodes on the key path too quickly, an energy level-based routing algorithm is put forward. The energy level-based routing algorithm will choose the path which has the highest energy level to deliver sample data to the sink nodes. The analysis shows that the minimum energy consumption routing algorithm adapts for data collection of sudden events and the energy level-based routing algorithm adapts for continuous query. Experiment results show that the energy level-based routing algorithm can keep the balance of energy consumption in the sensor network which can prolong the lifetime of the network.