Social interactions may encourage the cooperative behaviours by triggering either self-image concerns (when one sees others’ decisions without being seen) or social-image concerns (when one’s decision is seen by others). A laboratory experiment is designed to compare these two concerns directly, using a four-players finitely repeated public goods experiment on two directed star networks, self-image and social-image networks. The comparison of the players voluntary contributions in both types of networks reveals that their contributing behaviours are statistically indistinguishable. However, the players who belong to the self-image network are more willing to conform with the group behaviours, meaning that they will increase (reduce) the contributions if theirs are below (above) their groups average. Furthermore, I also find evidence that the contributing behaviours are more stable in the self-image networks than in the social-image network.