Given a network in which nodes are associated with a list of attributes, how can we define and characterize communities? How can we spot anomalous communities and anomalies within communities? Networks have long been studied and focus has most recently shifted to 'networks with content'. Long-studied network questions, such as ranking, clustering, and similarity, are reconsidered for such networks, as the new information such as node/edge attributes and types help enrich the formulations and increase our understanding of real-world networks. In this talk, I will introduce our work on spotting anomalies in networks with node attributes. Our main approach to anomaly mining in attributed networks is through communities. In particular, we quantify the degree that a community can be characterized through (a subset of) attributes on which its members 'click'. We then use such a quantity as a 'normality' score, based on which we identify individual anomalous nodes inside communities as well as communities that are anomalous as a group of nodes due to their low normality.