When studying the sampling methods on online social networks, samples collected by acceptance-rejection method are usually used as the "ground truth" to estimate the pros and cons of other sampling methods. The acceptance rate of the original acceptance-rejection method slumps dramatically since OSN sites updated their user ID from 32bit to 64bit. According to the characteristics of online social networks and taking Sina Weibo for example, we analyzed the distribution of user IDs in Sina Weibo, and proposed an improved acceptance-rejection method called UNI64. In this method, the user ID space is divided into valid intervals and vacant intervals by analyzing the distribution of valid sample IDs and utilizing cluster method. The sampling method generates candidate IDs only in valid intervals, so that the acceptance rate could be effectively improved even in a sparse-distributed user ID space.