An approximate algorithm of Top-k query based on sampling and weight in wireless sensor network was presented. The algorithm divides the network into several disjoint clusters in the sink node and the nodes in cluster to take sampling process. In the process of sampling, greater weight for reliable and important sensor node is given. The sensor node sensing data has a time correlation, and sampling threshold filtering in the cluster. Each cluster head node receives a Top-k candidate subset of the cluster, and then sends the subset to the sink node. Finally, the sink node can receive a Top-k sample candidate that represents the whole network. Simulation experiments show that the algorithm only needs to send small data and smaller samples, and can satisfy arbitrary precision requirements.