This paper explores the validity of social links as indicators of real user interactions in social networks, particularly in the context of socially-enhanced applications. The authors conduct a detailed study of user interactions on Facebook, the most popular social network in the US, to address the question of whether social links accurately reflect meaningful relationships. They propose the concept of *interaction graphs* to quantify user interactions and analyze their properties. The study finds that interaction graphs exhibit significantly different characteristics compared to social graphs, such as lower clustering coefficients and larger network diameters. These differences are quantified through various metrics, including degree distribution, clustering coefficient, and assortativity. The authors also validate their findings by comparing the performance of two socially-enhanced applications, Reliable Email (RE) and SybilGuard, on both social and interaction graphs. The results show that the choice of graph model significantly impacts the performance of these applications, highlighting the importance of using real indicators of user interactions rather than social graphs alone.This paper explores the validity of social links as indicators of real user interactions in social networks, particularly in the context of socially-enhanced applications. The authors conduct a detailed study of user interactions on Facebook, the most popular social network in the US, to address the question of whether social links accurately reflect meaningful relationships. They propose the concept of *interaction graphs* to quantify user interactions and analyze their properties. The study finds that interaction graphs exhibit significantly different characteristics compared to social graphs, such as lower clustering coefficients and larger network diameters. These differences are quantified through various metrics, including degree distribution, clustering coefficient, and assortativity. The authors also validate their findings by comparing the performance of two socially-enhanced applications, Reliable Email (RE) and SybilGuard, on both social and interaction graphs. The results show that the choice of graph model significantly impacts the performance of these applications, highlighting the importance of using real indicators of user interactions rather than social graphs alone.