A method of spatio-temporal channel assignment based on graph decomposition was proposed to solve the ripple effect and link failure problem in multi-radio wireless mesh networks. Network topology was decomposed into subgraphs in several time slots to extend channel assignment from spatial domain to time domain. Then, graph coloring algorithm was used to optimize the channel assignment on the basis of the subgraph in each time slot. Each subgraph obtained channel resource dynamically according to the network constraints to improve the channel assignment efficiency. Simulation results show that, compared to static channel assignment scheme, the impact of negative factors, such as the ripple effect and link failure caused by channel switching, are effectively confined with this strategy. The network throughput is increased at least by 30% while satisfying the constraints on the number of radios and the number of network channels.