Abstract: To research the influence of memory effect on online uses' behavior based on social network, the paper studied the records of some university users' access to social network and used the auto-correlation coefficient method to quantify the memory effect. Empirical research shows that the number of events obeys power-law distribution in the individual and group level. Therefore, the memory effect depends on the adjacent time interval. The continuous time interval less than some threshold are defined as burst periods. The result shows that the ratio of two adjacent time interval in a same active period shows the double Pareto distribution. According to the ratio, users increase or decrease the frequency of landing. The empirical data is consistent with the theoretical analysis. Thus, it is proved that there exists memory in the time interval.