A visual search is implemented when the eye moves to find a target symbol amongst many other symbols (distractors). The efficiency of a visual search is described by Hick’s law, which shows that the search time increases logarithmically as the number of symbols increases. In this paper, the efficiency of visual search was analyzed from the perspective of the network features of a conceptual ‘unobservable’ gaze position network superimposed on a monitor screen filled with many symbols (search array board). We assume that the gaze position does not move freely around the search array board, but rather moves in a way restricted to the unobservable gaze position network. First, we statistically verified that the artificial gaze position network designed from the data of visual search experiments have small-world features, and depends on the ratio of the saccades. Second, by implementing gaze step simulations on such small-world networks, we statistically verified that the simulation search times were close to those obtained from the experiments and also to the minimum search times. Thus, this study suggests that an efficient visual search can be explained by a small-world architecture hidden in the unobservable gaze position network and thus has to be artificially designed.