Due to such characters as diversity, low-speed, randomicity and self-organization of pedestrian flow, it is of great difficulty to study the behavior of pedestrians. This paper proposes a new method to study the bi-direction pedestrian flow by applying cellular automata (CA) combined with complex network theory. This paper designs the survey experiment to study the features of pedestrians' walking preference. Then the cellular automata model considering pedestrians' walking preference features is built in which the forward-parameter, right-parameter, surpass-parameter and the surrounding-correction parameters are brought to mend the transition probability. Based on the k-Nearest-Neighbor interaction pattern, the bi-direction pedestrian flow of the CA model is abstracted as complex network of pedestrians. The simulation results show the phase transition, and density-speed, density-volume curves of pedestrian flow. At the same time, the parameter of pedestrian complex network is generated. Then, the correlation between the average speed of the bi-direction pedestrian flow and the average path length as a parameter of the network's structure characteristic is found, that is the pedestrian flow with shorter average-path length operates with higher average speed.