How corruption behavior develops？The theoretical derivation implies that corruption should decrease as time pass by. Because if the probability of being caught is p every time, then after n times of corruption, the probability of being caught is increasing exponentially to 1-(1-p)n . Contradictorily, the practical experience indicates that people often increase corruption. There could be two explanations to such contradiction. Firstly it may because of the ignorance of the real corruption development trends for those corruptors who didn’t get caught nor exposed. Secondly, it may because of the negligence of some key subjective factors in theoretical derivation, like perceived probability and controllability of risk. Therefore, the study simulated a corruption situation in laboratory to (1) Explore how corruption behavior develops with time passing by, (2) Explore the effects of perceived probability and controllability of the risk on people’s corruption behavior. Participants were 155 college students in Beijing (37% males). We simulated a common bribery situation in our laboratory, and observed how people behave in 10 rounds corruption simulated situations. The participants played as government officials, and faced with bribe from faked businessman. They needed to decide whether to accept it or not. And if they accepted the bribe, based on the amounts of bribe, there was a certain probability they would get punished. However participants could avoid the punishment by finishing a task. We manipulated the level of risk probability and risk controllability in the corruption situations. Using latent growth mixture modeling, three distinct linear growth patterns of corruption behavior were identified. First class (39.36%), with a middle initial status of corruption, demonstrated a slow non-significant linear decrease in the following times. We call this class “normal-corruptors”. The second class (30.97%), with a higher initial status of corruption, demonstrated a slowly non-significant linear increase in the following times. We call this class “slowly-falling-corruptors”. The last class (29.68%) remained a very low degree of corruption, or no corruption at all. We call this class of people “non-corruptors”. Analyses also showed that the probability and controllability of risk have different effects on corruption development. The probability of risk only has effect on the initial status of corruption. However the controllability of risk has effect not only on the initial status but also on the slope of the developmental trends. In the condition of a low-level risk probability, people intended to have a higher initial status of corruption compared to the condition of a high probability. And in the condition of a high-level controllability, people intended to have a higher initial status of corruption and also they tended to have a more rapid development in the following times. The results suggested that the corruption trajectory is heterogeneous and there are three different classes of corruption development trajectories. The research also suggested that increasing the risk probability of corruption is not as effective as we thought before. And in order to keep corruption within limits, we need to pay more attention to reduce the people’s controllability.