In order to investigate the following behaviors of pedestrians under limited visibility, pedestrian evacuation is simulated by an improved cellular automata model. Considering the familiarity of pedestrians with the environment, pedestrians are divided into two types: Informed type and uninformed type. For the informed pedestrians, an extended cost potential field cellular automata model is proposed. For the uninformed pedestrians, some following behavior strategies are suggested to study their evacuation behaviors. These following behaviors include following a pedestrian in the visibility (S1), following the most people’s position in the visibility (S2), following the most movement direction in the visibility (S3), and walking along the wall (S4). To investigate the evacuation efficiency of these different following strategies, we compare the performances of different densities of informed pedestrians, different visibility and different pedestrian proportions. As demonstrated by the simulation results, evacuation efficiency and the effectiveness of the following strategies are related to the visibility and density of informed pedestrians. The simulation results show that when the density of informed pedestrians is constant, S4 is more efficient at very poor visibility (R = 1), strategy S3 and S4 are more efficient at poor visibility (1 < R < 7), and the four strategies are equally efficient at good visibility ( R > 6). In addition, when the visibility is constant, the density of informed pedestrians is less than 0.5, strategy S3 is the most efficient strategy. When the visibility is constant, the density of informed pedestrians is more than 0.5, the four strategies have a better performance. Moreover, it is noted that the same regular changes also exist in a single strategy environment. These findings can provide some insights into the emergency evacuation of large public places such as supermarkets and stadiums, and help develop effective guidance strategies under limited visibility.