The requirement for measuring the similarity between web pages arises in many applications on the Web, such as web searching engine and web document classification. According to the unique characteristics of the Web, which are huge, rapidly growing, high dynamic, and untrustworthy, we propose a novel link-based similarity measure called PageSim. Based on the strategy of PageRank score propagation, PageSim is efficient, scalable, stable, and "fairly" robust, and therefore is applicable to the Web. We present intuitions behind the PageSim model, and outline the model with mathematical definitions. We also suggest the pruning technique for efficient computation of PageSim scores, and conduct experiments to illustrate the effectiveness and specialities of PageSim.