Abstract: Information flow network, which was constructed on information flows between stocks traded in China's stock market over the period from 2005 to 2016, was studied. Transfer entropy was employed to quantify asymmetric and non-linear information flows between stocks. Then directed-weighted information flow networks were constructed by employing a threshold value and sliding window method. Average path length and clustering coefficient borrowed from complex theory and methods such as intentional attack and random failure were used to analyze the characteristics of networks, such as network topologies, key nodes and small-worldness. Studies showed that there was intense information interactions between stocks during both volatile and relative stable periods in question, and structural differences are observed among time-varying information flow networks. Changing characteristics of network topologies show that information flow networks at certain calendar dates showing effects of small-world and with presence of nodes exhibiting extreme influences at the same time. Furthermore, small-worldness analysis based on intentional attack and random failure imply that risk control based on key stocks will reduce global risk but may intensify local risk contagion.