Cooperation in social dilemmas has received considerable attention during the last decades. In repeated social dilemmas settings, individuals’ contributions decline over time and finally reach a minimum when the interaction terminates. However, it is observed that individuals try to increase their contributions in their initial stage of interactions. A previous study using a step-level public good dilemma has verified that those who contributed high in their first interaction demonstrated high levels of contribution in their subsequent interactions, whereas those who contributed low increased their levels of contribution subsequently. In addition, the cooperative level of low contributors increased along with their cooperative belief. In current study, we aimed to replicate those findings using a different task, i.e., a continuous public good dilemma, and to examine the role of cooperative belief in cooperation. The design was a 2 (peer behavior: equality violation vs. equality compliance) × 2 (decision stage: first vs. second) mixed factor design, with the latter being a within-subject variable. 60 college students (Mage=21.13, 27 males) were recruited from a university campus. They were randomly assigned to the equality compliance or violation condition. For each session, a group of six participants were first instructed to make their allocations in a continuous public good game, and then assessed their cooperative belief. Thereafter, manipulation on peer behavior was introduced. In the equality compliance condition, participants was believed that most members complied with an equal allocation rule, whereas in the equality violation condition, only one of 5 other members gave a cooperative allocation while the rest contributed far less than an equal level. Thereafter, the participants were provided another chance to make their allocation decision and to rate their cooperative belief. In addition, they also assessed their social value orientation. Based on their first decisions, participants were divided into two groups, i.e., low-contributors (N=23) who allocated less than 50 tokens to the group, and high-contributors (N =37) who allocated more than 49 tokens to the group. Our result did not support that this classification was related to participants’ reliable social value orientation. A 2 × 2 × 2 (peer behavior × participant type × decision stage) ANOVA on contributions revealed participants in the equality compliance condition cooperated more than those in the equality violation condition. In addition, participants increased their allocations from their first interaction to their second interaction. Importantly, the high contributors demonstrated high levels of contribution in their subsequent interactions, whereas the low contributors increased their levels of contribution after the initial interaction because of their increase of cooperative belief. In sum, the present study replicates the previous finding, that is, people increased their contributions during initial social interactions. This may suggest that people are in nature other-regarding. However, the observed increase in the cooperative level is due to individuals who contribute little in their first interaction. Therefore, one practical implication is that individuals who contribute less in the initial interactions should garner greater attention than those who contribute more.