Abstract: The limited amount of available orthogonal channels and noise interference are two major issues to constrain performances of multi-channel MAC protocols in wireless sensor networks (WSN). In conjunction with features of data gathering networks, a partitioning-based multi-channel MAC protocol was designed. Initially it partitioned network into several sub-trees with lower collision factor, without increasing the total value of interference around the whole network. Furthermore, it adopted a coarse-grained channel assignment strategy on basis of graph coloring theory, in order to ensure each sub-tree work on a relative optimal channel. Simulation results show that the proposed MAC protocol can significantly improve throughput and reduce end-to-end latency and packet loss.