The explosive growth of online social networks in recent years have generated massive amount of data-sets in user behaviors, social graphs, and contents. Given the scale, heterogeneity, and diversity of such big data, sampling becomes a simple and intuitive approach to reduce the size of the data-sets for collecting, measuring, and understanding users, behaviors and traffic in online social networks. In this paper, we quantify the impact of random sampling on the analysis of online social networks with Twitter streaming data as a case study. In addition, we design different sampling strategies including community sampling and strata sampling, and evaluate their impact on a broad range of behavioral characteristics of online social networks. Our experimental results show that community sampling has the minimum impact on tweet distributions across users and the structure of retweeting graphs, while achieving the similar data reductions as random and stratified sampling.